The gut metabolite indole-3 propionate promotes nerve regeneration and repair

The regenerative potential of mammalian peripheral nervous system neurons after injury is critically limited by their slow axonal regenerative rate1. Regenerative ability is influenced by both injury-dependent and injury-independent mechanisms2. Among the latter, environmental factors such as exercise and environmental enrichment have been shown to affect signalling pathways that promote axonal regeneration3. Several of these pathways, including modifications in gene transcription and protein synthesis, mitochondrial metabolism and the release of neurotrophins, can be activated by intermittent fasting (IF)4,5. However, whether IF influences the axonal regenerative ability remains to be investigated. Here we show that IF promotes axonal regeneration after sciatic nerve crush in mice through an unexpected mechanism that relies on the gram-positive gut microbiome and an increase in the gut bacteria-derived metabolite indole-3-propionic acid (IPA) in the serum. IPA production by Clostridium sporogenes is required for efficient axonal regeneration, and delivery of IPA after sciatic injury significantly enhances axonal regeneration, accelerating the recovery of sensory function. Mechanistically, RNA sequencing analysis from sciatic dorsal root ganglia suggested a role for neutrophil chemotaxis in the IPA-dependent regenerative phenotype, which was confirmed by inhibition of neutrophil chemotaxis. Our results demonstrate the ability of a microbiome-derived metabolite, such as IPA, to facilitate regeneration and functional recovery of sensory axons through an immune-mediated mechanism.

Injuries to the peripheral nervous system (PNS) have high prevalence and are typically followed by long-lasting neurological disability without effective treatments beyond surgical reconstruction, which is effective only in a small percentage of cases [6][7][8] . These injuries frequently result in partial or total loss of sensory, motor and autonomic functions owing to inefficient and slow axonal regeneration 9,10 . Because a delayed target re-innervation leads to irreversible loss of function of target organs 11 , accelerated axonal regeneration is required to enhance functional outcomes following injury. Studies of cellular and molecular responses to injury 12,13 , neurodevelopmental pathways 14,15 , regenerative organisms 15,16 and molecular screening approaches 17 have contributed to the identification of numerous mechanisms that influence regenerative ability. However, our knowledge of the cellular and molecular mechanisms underpinning axonal regeneration and functional recovery remains incomplete, which undermines the development of effective treatments for clinical injuries.
Recently, accumulating evidence has implicated healthy lifestyle choices such as exercise and environmental enrichment in priming and enhancing the regenerative potential of sensory neurons 3 . Similarly, dietary regimens including intermittent fasting (IF) have been shown to promote metabolic and signalling pathways with the potential to favour wound repair, neuronal sprouting and recovery in several disease states, including that occurring after nervous system injury [18][19][20][21] . Furthermore, IF has been shown to increase synaptic plasticity 22,23 and neurogenesis 24 , partially sharing molecular mechanisms with axonal regeneration. Therefore, we investigated whether IF would prime sensory neurons for enhanced axonal regeneration and recovery in a model of sciatic nerve injury and whether it would enable the identification of specific molecular mechanisms underpinning nerve repair.

IF promotes regeneration through the gut microbiome
Axonal regeneration was initially assessed 24 h after a sciatic nerve crush (SNC) that occurred 10 or 30 days after initiation of an IF versus an ad libitum (AL) diet. The 10-day and 30-day IF regimens significantly enhanced regeneration past the crush site to a similar extent (Extended Data Fig. 1a-c); therefore, all subsequent experiments were performed after 10 days of IF. Notably, IF also promoted axonal regeneration when the recovery time was extended to 72 h after SNC (Fig. 1a-c). In confirmation of these findings, IF was followed by an increase in neurite outgrowth in cultured dorsal root ganglion (DRG) neurons (Extended Data Fig. 1d-f). Notably, IF did not alter critical modifiers of axonal regeneration such as Schwann cell and macrophage recruitment 25,26 to the nerve crush site (Extended Data Fig. 1g-j) or the neurotrophic factors BDNF, NGF, NT-3 and NT4/5 in the DRG [27][28][29][30][31] 72 h after SNC (Extended Data Fig. 1k). These results indicate that Schwann cells, macrophages and neurotrophic factors may not have a central role in the IF-dependent regenerative phenotype.
Given that IF modifies cell metabolism 32,33 , we considered that changes in metabolism may underpin the IF-dependent regenerative ability. To test this, semi-targeted gas chromatography coupled to mass spectrometry (GC-MS) was performed for serum extracted after mice underwent IF versus AL. This enabled us to identify 79 metabolites, 14 of which were differentially enriched following IF ( Fig. 1d-f, Extended Data Fig. 2a-c and Supplementary Data 1). These metabolites could be clustered into two groups: microbiome-derived metabolites and host metabolites (Fig. 1e). Interestingly, the four most enriched metabolites following IF were the microbiome-derived 3-indolelactic acid 2, 2,3-butanediol 2, xylose 2 and indole-3-propionic acid (IPA; Fig. 1f). To assess whether the IF-dependent microbiome promotes sciatic nerve regeneration, microbiota collected from IF-animals were transplanted into AL-mice through faecal microbiota transplantation (FMT). Indeed, FMT from the IF-mice led to a significant increase in regenerated fibres past the crush site in AL-mice after SNC (Extended Data Fig. 2d-f). Because indole metabolites are synthesized primarily from tryptophan by the gram-positive bacteria Bifidobacterium (indole-3-lactic acid) 34 , Lactobacillus and Clostridium sporogenes (IPA) 35    The dashed line indicates the crush site. c, Quantification of the percentage of fibres past the crush site normalized to the number of fibres at the crush site plotted as a function of the distance from the crush site (n = 4 biologically independent animals per group with a bilateral SNC; ****P < 0.0001, two-way analysis of variance (ANOVA) with Tukey's post hoc test examined over three independent experiments; mean ± s.e.m.)). d, Score plots for the first predictive (t [1]) and orthogonal (to [1]) components of the orthogonal partial least squares discriminant analysis (OPLS-DA) models based on metabolic profiles obtained by GC-MS (n = 10 biologically independent animals per group). e, Table listing significantly upregulated and downregulated metabolites in serum from mice that underwent IF versus an AL diet (P < 0.05). Red text indicates the top metabolites that can be produced by bacteria. Shown are the logarithmic fold changes (logFC) and the P values from comparison of the IF and AL groups (n = 10 biologically independent animals per group; PLS-DA was used to identify the significantly differential metabolites between the groups). f, Bar graphs plotting the relative abundance of gut microbiome-derived metabolites obtained from the GC-MS analysis, including IPA, 3-indolelactic acid 2, 2,3-butanediol 2 and xylose 2 (n = 10 biologically independent animals per group; two-tailed unpaired Student's t-test; mean ± s.e.m.). regeneration ( Fig. 2a-c), indicating that gram-positive bacteria are required for IF-dependent regeneration.
To identify metabolites underpinning the microbiome-dependent regenerative phenotype, serum samples were profiled by GC-MS after IF or AL with or without vancomycin treatment (Fig. 2d-k and Extended Data Fig. 2g-j). Notably, IPA was the only metabolite significantly affected by vancomycin treatment with the most significant interaction P value, which suggests that its amount critically depended on IF and the gram-positive gut microbiome ( Fig. 2e and Extended Data Fig. 2j). Indeed, after IF, serum IPA was found among the metabolites with the most increased abundance, and this increase was completely blocked by vancomycin (Fig. 2f), as confirmed by targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS; Fig. 2g,h). As expected, vancomycin significantly reduced the bacterial diversity, as shown by 16S rDNA sequencing (Extended Data Fig. 3a,b and Supplementary Data 2), and decreased the abundance of Bacteroidetes and Firmicutes, which were increased after IF (Extended Data Fig. 3c and Supplementary Data 3). Of note, the vast majority of Firmicutes belonged to the order Clostridiales (Extended Data Fig. 3d and Supplementary Data 3), and the five bacteria with the greatest increase in abundance after IF were also in the Clostridiales order (Extended Data Fig. 3e). This confirms that vancomycin treatment affects IF-dependent pathways. Next, we used Mice underwent IF or AL and were treated with vancomycin or vehicle for 10 days followed by SNC and 3 days of axonal regeneration. b, Micrographs of representative longitudinal sections of sciatic nerves for all groups, immunostained for SCG-10. Scale bar, 1,000 µm. The dashed line indicates the crush site. c, Quantification of the percentage of fibres past the crush site normalized to the number of fibres at the crush site plotted as a function of the distance from the crush site (n = 4 biologically independent animals per group with a bilateral SNC; ****P < 0.0001, two-way ANOVA with Tukey's test examined over two independent experiments; mean ± s.e.m.). d-k, GC-MS metabolomics analysis conducted from serum after 10 days of IF or AL with or without vancomycin treatment. d, Score plot of the PLS-DA analysis performed on normalized GC-MS metabolomic data (n = 10). e, Variable correlation analysis modelling the overlap between loadings and a score plot showing metabolites that were represented primarily in specific groups. f, Complete depletion of IPA by vancomycin (n = 8 (AL) and 9 (IF) biologically independent animals per group; two-way ANOVA with Tukey's multiple-comparisons test, P value indicates comparison to AL; mean ± s.e.m.). g, LC-MS/MS measurements of IPA in serum from mice following 10 days of IF versus AL (n = 5 biologically independent animals per group; two-sided Student's t-test; mean ± s.e.m.). h, LC-MS/MS measurements of IPA in serum from mice following 10 days of IF versus AL with or without vancomycin treatment (n = 3 biologically independent animals per group; two-way ANOVA with Tukey's multiple-comparisons test; P value indicates comparison to AL; mean ± s.e.m.). i-k, Relative abundance of 3-indolelactic acid, 2, 2,3-butanediol 2 and xylose 2 across experimental groups (n = 10 mice per group; two-way ANOVA with Tukey's multiple-comparisons test, P value indicates comparison to AL control group; mean ± s.e.m.). Piphillin software, which predict functional metagenomic data on the basis of annotated genome databases (Extended Data Fig. 4 and Supplementary Data 4). This analysis identified IF-dependent enrichment, which was reversed by vancomycin, for pathways that reflect IPA and IPA pregnane X receptor (PXR) function such as xenobiotic biodegradation, anti-microbial drug resistance and metabolism of cofactors and vitamins (Extended Data Fig. 4). Additionally, analysis based on the GeneWays system, which was used to predict gene abundance within each group, indicated IF-dependent enrichment for tryptophan synthase (TrpA), which converts indole-3-glycerol phosphate into indole (Supplementary Data 4). Notably, vancomycin abolished this enrichment. Together, these data indicate that gram-positive bacteria producing IPA may be responsible for IF-dependent axonal regeneration.

IPA enhances repair through neutrophil chemotaxis
Next, we investigated whether IPA synthesis by the gut microbiome is required for axonal regeneration following SNC. After   These results together demonstrate that bacterial production of IPA in the gut critically affects axonal regeneration. Considering that changes in IPA production in the gut are reflected by variations in the serum, we administered IPA by intraperitoneal (i.p.) injection, and the axonal regeneration was assessed 72 h after SNC. Notably, the axonal regeneration with systemic IPA delivery mirrored the increase in regeneration caused by IPA gavage (Extended Data Fig. 5f-h), indicating that serum IPA might affect DRG neuron growth either directly or indirectly through neuron intrinsic or neuron extrinsic mechanisms, respectively.
When IPA was administered directly into the media of cultured DRG neurons, no changes in neurite outgrowth were observed (Extended Data Fig. 5i,j). This suggests that IPA affects through a regenerative mechanism extrinsic to DRG neurons. We also found an increase in axonal regeneration when IPA was administered after an injury, as indicated by both immunolabelling of SCG-10-positive regenerating axons and dextran injections administered distal to the nerve crush site (Fig. 3i,j and Extended Data Fig. 5k-n).
To investigate IPA-dependent molecular signatures that potentially affect the regenerative ability of DRG neurons, we performed RNA sequencing on RNA from the DRG before (naive) or 72 h after SNC for mice that received IPA gavage (20 mg kg −1 per day) for 10 days before the injury and 3 days afterwards or PBS oral gavage. The IPA treatment affected the DRG gene expression programme, as indicated by differential gene expression analysis (Extended Data Fig. 6a-c). Expression of the gene encoding the IPA nuclear receptor PXR (Nr1i2) was increased (Extended Data 5), which supports the activation of IPA signalling. In particular, IPA affected genes related to immune-regulatory Gene Ontology (GO) categories, with the most significant enrichment in 'neutrophil chemotaxis' 72 h after SNC (Fig. 4a,b). Notably, Cd177 (neutrophil ligand) and Cxcl1 (endothelial neutrophil chemoattractant chemokine (CXC motif) ligand 1) were selectively upregulated by IPA after SNC (Extended Data Fig. 6c), suggesting that IPA might promote neutrophil chemotaxis towards the DRG through a CXCR2-mediated mechanism. In support of this model, we found that IPA treatment increased the number of neutrophils within the DRG tissue but not at the nerve crush site (Extended Data Fig. 6d-k and Extended Data Fig. 7). The number of other immune cells, including CD4 + and CD8 + T cells, macrophages, natural killer (NK) cells and B cells, did not differ between the mice receiving IPA and vehicle (Extended Data Fig. 8). Neutrophil depletion with anti-Ly6G monoclonal antibody significantly decreased the axonal regeneration mediated by IPA compared with the IgG control antibody (Extended Data Figs. 7 and 9a-k). Notably, PXR deletion reduced the number of neutrophils and IPA-dependent regeneration to a similar extent the anti-Ly6G monoclonal antibody (Extended Data Fig. 9j-n).
DRG immunolabelling showed that the majority of neutrophils were positive for the CXCL1 receptor CXCR2 (Extended Data Fig. 10a,b), which is a marker of neutrophil activation and chemotaxis. This suggests that neutrophils might be activated and attracted by a CXCR2-mediated mechanism.
To test this hypothesis, neutrophil chemotaxis was inhibited at the time of IPA treatment by using an anti-CXCR2 neutralizing monoclonal antibody, and this significantly impaired the IPA-dependent regeneration after SNC ( Fig. 4c-g and Extended Data Fig. 10c,d). Taken together, these data indicate that IPA-dependent neutrophil chemotaxis is required for axonal regeneration after SNC.
RNA sequencing also identified an IPA-dependent increase in interferon gamma (IFNγ) signalling in the DRG, indicating a potential signalling mechanism downstream of neutrophil recruitment (Extended Data Fig. 11a). Indeed, monoclonal antibody-mediated neutralization of IFNγ significantly reduced IPA-dependent sciatic nerve regeneration (Extended Data Fig. 11b,c). Further, in vivo administration of IFNγ led to increased regenerative growth of cultured DRG neurons (Extended Data Fig. 11d,e). Similarly, IFNγ treatment in cultured DRG neurons induced a significant increase in regenerative growth that was blocked by anti-IFNγR neutralizing antibody (Extended Data Fig. 11f,g). Notably, IFNγR was expressed in DRG neuronal soma but not in the axons (Extended Data Fig. 11h,i), and in vivo IFNγ delivery alone increased the expression of several regenerative signals in the DRG (Extended Data Fig. 11j,k).

IPA improves recovery and skin innervation
Next, we investigated whether IPA administration at a clinically suitable time 24 h after sciatic nerve injury is able to accelerate the rate of sensory Shown are significantly upregulated and downregulated GO pathways categorized by the Database for Annotation, Visualization, and Integrated Discovery (DAVID) Bioinformatics Resource with −log 10 (P) < 0.05. The statistically significant (modified Fisher's exact P ≤ 0.05) GO categories highlighted in red and green are upregulated and downregulated pathways, respectively. c, Ly6G immunostaining in sciatic DRG sections of mice treated with IPA or PBS and anti-CXCR2 or anti-IgG monoclonal antibody 3 days after SNC. Scale bar, 50 µm. d, Quantification of the number of Ly6G + cells in sciatic DRG sections of mice treated with IPA or PBS and anti-CXCR2 or anti-IgG monoclonal antibody 3 days after SNC (n = 4 independent experiments per group; one-way ANOVA with Sidakʼs multiple-comparisons test; mean ± s.e.m.). e, Schematic of the experiment presented in f and g. Mice were treated for 10 days with either IPA (20 mg kg −1 per day) or PBS and anti-IgG or anti-CXCR2, followed by SNC and 3 days of axonal regeneration time. f, Micrographs of representative sections of sciatic nerves immunostained for SCG-10. Scale bar, 1,000 µm. The dashed line indicates the crush site. g, Quantification of the percentage of fibres past the crush site at the indicated distances normalized to the number of fibres at the crush site (n = 4 biologically independent animals per group with bilateral SNC; ****P < 0.0001, two-way-ANOVA with Sidakʼs test; mean ± s.e.m.). h, i, Schematic of the experimental design. j, Visualization of the paw withdrawal latency in seconds for the Hargreaves test for thermal pain sensation (n = 10 biologically independent animals per group; two-way ANOVA with Sidakʼs test; mean ± s.e.m.). k, Von Frey analysis for nociception (stimulus intensity shown in grams), with no significant difference between the groups when plotted in a timeline (n = 12 biologically independent animals per group; two-way ANOVA with Sidakʼs multiple-comparisons test; mean ± s.e.m.; NS, not significant). l, PGP9.5 immunostaining with 4′,6-diamidino-2-phenylindole (DAPI) counterstaining showing epidermal innervation of the hindpaw interdigital skin 16 days after SNC. The dashed lines indicate the boundary between the epidermis and dermis. Scale bar, 100 µm. m, n, Quantification of the percentage of intra-epidermal nerve fibres (IENF) versus dermal nerve fibres (DNF) and the number of IENF per millimetre of interdigital skin after PBS or IPA treatment (n = 4 biologically independent animals per group, two-sided Student's unpaired t-test).
recovery following the injury. In this analysis, we evaluated both the recovery of physiological sensory function by assessing responses to thermal stimulation and the presence of mechanical allodynia, which can be an undesired consequence of maladaptive epidermal re-innervation ( Fig. 4g-j). In the Hargreaves test, IPA-treated mice recovered more quickly from thermal nociception, showing a reduced withdrawal response time compared with PBS-treated mice between day 7 and day 19 (Fig. 4i). However, no differences in mechanical allodyniano, as measured by Von Frey filaments, were observed between the groups treated with IPA and PBS (Fig. 4j). These data show that IPA leads to accelerated recovery in thermal nociception, but it does not induce mechanical allodynia following sciatic nerve injury. Notably, anti-PGP9.5 immunostaining of the hind paw interdigital skin 16 days after SNC indicated a significant increase in epidermal innervation following IPA versus PBS treatment, providing an anatomical substrate for functional recovery (Fig. 4k-m). These results together indicate that IPA may induce sensory recovery by improving epidermal innervation without causing neuropathic pain in a sciatic crush model of peripheral nerve injury.

Conclusions
The present study identifies a dietary-dependent regenerative mechanism that relies on production by gut gram-positive Clostridiales bacteria of the gut microbial metabolite IPA. IPA promotes axonal regeneration, epidermal innervation and sensory neurological recovery after SNC through the recruitment of neutrophils to the DRG. IPA, a product of bacterial metabolism of tryptophan in the lower gut, and its receptor PXR have been implicated in a variety of immune responses in several tissues and cell types 37-41 , including in the gut microenvironment, where immune suppressive and anti-inflammatory functions for IPA and PXR signalling have been described 37-39 . Conversely, PXR activation in vascular endothelial cells has been shown to initiate an innate inflammatory response 41 , which results in neutrophil recruitment 42 . This supports our model suggesting an IPA-dependent increase in transendothelial neutrophil chemotaxis.
PXR expression has recently been found in immune cells 40,43,44 . One study demonstrated that PXR ligand binding in macrophages activates an innate inflammatory response, leading to activation of interleukin (IL)-1β and NRLP3 (ref. 40 ). Similar results were found in vascular endothelial cells 41 . In line with this, several genes summarized as innate immune response genes, specifically in neutrophil/IFNγ signalling, were upregulated in our dataset following IPA treatment before and after an injury. This indicates injury-independent activation of immune signalling by potential PXR activation. Additional important experiments comparing WT with PXR knockout (KO) mice allowed us to determine that IPA-dependent nerve regeneration requires PXR and that PXR KO mice do not display an increase in neutrophils in the DRG following IPA administration.
Neutrophils are required for axonal regeneration in the PNS 45 and central nervous system 46-48 via changes in their recruitment at the lesion site. Our study supports an additional regenerative function for neutrophils by communicating regenerative signals such as IFNγ at the level of DRG where the neuronal cell bodies reside.
IFNγ-dependent axonal regeneration is in line with increased inflammatory signalling by neutrophils. However, a role for additional cytokines supporting IPA-dependent axonal regeneration cannot be ruled out.
In humans, the gut microbiome undergoes profound changes when the nervous system is injured 49 . Accordingly, identifying gut microbiome-derived regenerative small molecules for oral or intravenous administration could provide promising new therapeutic possibilities for nerve injury, promoting axonal regeneration and enhancing neurological recovery.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04884-x. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Animal husbandry
Animal work was carried out in accordance with Home Office under the Animals (Scientific Procedures) Act 1986, with local ethical review by the Imperial College London Animal Welfare and Ethical Review Body Standing Committee (AWERB). Mice were maintained in house under standard housing conditions (12-h light/dark cycles) with 24-h access to water and a standard chow diet at temperatures between 20 °C and 24 °C and relative humidity between 45% and 65%, according to the UK Home Office Code of Practice. All WT mice were of a C57BL/6 background (male, 20-30 g, 6-8 weeks of age).
The generation of PXRKO mice has been described previously 50 . The PXRKO sperm (kindly provided by the laboratory of S. Mani, Albert Einstein College of Medicine) was rederived in C57BL/6 embryo donors. All PXRKO and PXRWT mice used in experiments were male, 20-30 g and 6-8 weeks of age.
When the mice were killed, anaesthesia by isoflurane inhalation or i.p. injection of ketamine (100 mg kg −1 )/xylazine (10 mg kg −1 ) was used to minimize suffering of the animals. Cervical dislocation was performed following anaesthesia to ensure death.
Experimenters were blinded to experimental groups during scoring and quantifications. Animals were randomized in all experiments. Sample sizes were chosen in accordance with similar experiments published previously.

IF
Body weight-matched male C57BL/6 mice of 6-8 weeks old were randomly assigned to IF or ad libitum (AL; control) treatment groups. The IF group did not have access to food during the first 24 h; afterwards, food was withheld every second day. Specifically, fasting occurred 0-24 h, 48-72 h and 96-120 h after the start of the experiment, and the AL diet was available on the alternating days, 24-48 h, 72-96 h and 120-144 h after the experiment began. Pre-weighed food was provided in the food hopper of the home cage at 09:30 local time except for fasting days for the IF groups, and the uneaten food was weighed 24 h later.

SNC surgery
Mice were anaesthetized with isoflurane (5% induction, 2% maintenance), shaved on the hind limbs and lower back and sterilized with iodine, and an ophthalmic solution was applied to the eyes to prevent drying. An incision was made on the skin and the biceps femoris, and the gluteus superficialis was opened by blunt dissection. The sciatic nerve was then exposed using a surgical hook. The SNC was performed orthogonally for 20 s (45 s for the reinnervation experiment) using 5-mm surgery forceps (91150-20, Inox Electronic) followed by carbon crush-site marking. The crush was performed approximately 20 mm distal to the sciatic DRG.
For dextran tracing, immediately following the crush, 2 µl of dextran was injected 2 mm distal to the crush site using a 1-mm Hamilton syringe. The DRG was dissected 72 h after surgery.

FMT
FMT was carried out as described previously 51,52 . In detail, donor and recipient C57BL/6 mice (SPF) were co-housed in groups of four per cage. Donor mice underwent either 10 days of IF or AL (control condition) before the FMT start. The recipient animals were pre-treated for 3 days with vancomycin (V2002, Sigma; 50 mg kg −1 per day, 214.27 mg l −1 ) administered through drinking water before the FMT, and they had access to the AL diet throughout the experiment. FMT was carried out for IF or AL mice daily for 10 days as follows. Fresh faeces were collected at 09:00 local time from the cage of five healthy C57BL/6 mice (IF or AL). The mice were placed in a clean cage, and 1 g of faeces was collected using sterile forceps. The pooled faeces were then homogenized in 10 ml of PBS using a vortex followed by centrifugation for 30 s at 3,000 r.p.m. at 4 °C. Afterwards, the supernatant was transferred to a new tube and was centrifuged for 5 min at 12,000 r.p.m. at 4 °C. Finally, the bacterial pellet was resuspended in 2.5 ml of PBS. At about 12:00 local time, the FMT was then administered by oral gavage at a dosage of 500 µl for each mouse.

Antibiotic administration
Vancomycin (V2002, Sigma) was administered through the drinking water (50 mg kg −1 per day; 214.27 mg l −1 ) and was replenished every second day.
IPA treatment IPA (57400, Sigma) was diluted in sterile PBS at a concentration of 0.5 mg per 200 µl. The mice were treated with 10 or 20 mg kg −1 per day by gavage or i.p. injection.
C. sporogenes recolonization C.s. WT and C.s. fldC mutants were cultured anaerobically overnight in trypticase yeast extract medium. Bacterial cultures were mixed with glycerol in hermetically sealed glass vials to obtain final concentrations of 6.01 × 10 6 CFU ml −1 for C.s. WT and 5.46 × 10 6 CFU ml −1 for C.s. fldC . The mice were pre-treated for 3 days with vancomycin administered in the drinking water, and the C.s. WT or C.s. fldC strains were transplanted into the caecum by oral gavage for 10 consecutive days at 1 × 10 6 CFU per day.

Serum and caecum preparation
Blood was extracted by heart puncture and collected into a covered test tube. The blood was allowed to clot for a minimum of 30 min at room temperature and was then centrifuged at 2,000 g for 10 min at 4 °C. The supernatant was collected into a fresh tube and stored at −20 °C until use.
Caecum content was removed from the caecum using sterile forceps, placed into a tube and immediately snap-frozen in liquid nitrogen.

GC-MS untargeted metabolomics
The serum samples (100 µl) were prepared as follows. The samples were spiked with 10 µl internal standard (IS) solution (myristic acid-d27, 750 mg ml −1 , and 850 µl of ice-cold methanol was added, followed by centrifugation for 20 min (4 °C, 16,000g). Then, 750 µl of supernatant was transferred to silanized dark 2 ml autosampler vials and was evaporated to dryness in a rotational vacuum concentrator (45 °C, 20 mbar, 2 h). Next, 50 µl of methoxyamine solution (2% in pyridine) was added, and the samples were incubated overnight at room temperature. Afterwards, 100 µl of N-methyl-trimethylsilyl-trifluoroacetamide (MSTFA) + 1% trimethylchlorosilane (TMCS) solution was added. The samples were incubated at 60 °C for 1 h and were transferred to dark autosampler vials with 250-µl silanized inserts. The samples were analysed in an Agilent 7890B-5977B Inert Plus GC-MS system, with 2 µl of each sample injected in a split inlet (1:10 ratio). The chromatographic column was an Agilent ZORBAX DB5-MS (30 m × 250 µm × 0.25 µm, 10 m Duraguard). The temperature gradient was 37.5 min long, and the mass analyser was operated in full-scan mode at 50-600 m/z. The detailed instrumental conditions can be found in the G1676AA User Guide available at the Agilent Fiehn GC/MS Metabolomics RTL Library at https://www.agilent.com/cs/library/usermanuals/Public/ G1676-90001_Fiehn.pdf. Quality control (QC) samples were created by pooling equal amounts of each serum sample and were analysed interspaced in the analytical run. The study samples were randomized before sample preparation. Peak deconvolution, alignment and annotation were performed for the pooled QC samples using software packages AMDIS (NIST), Mass Profiler Professional and Unknowns Analysis (Agilent Technologies) available at FiehnLib at https://fiehnlab.ucdavis. edu/projects/fiehnlib. Peak picking was performed with GAVIN (https:// doi.org/10.1016/j.ab.2011.04.009) 53 . Non-reproducible (coefficient of variation (CV) > 30% in QC samples) and contaminated (blank > 20% of the mean QC levels) metabolic features were removed from the dataset.
This dataset was then used to build orthogonal partial least squares discriminant analysis (OPLS-DA; IF versus AL) or partial least squares discriminant analysis (PLS-DA; IF, IF-vancomycin (IF-V), AL, AL-vancomycin (AL-V)), focusing on the differences among the experimental groups using R statistical environment version 3.4.4 (ropls R package). The OPLS algorithm is derived from the PLS regression method 54 , which explains the maximum separation between class samples y (n dummy variables for n classes) by using the GC-MS data x. Here the ropls R package (http://bioconductor.org/packages/ release/bioc/html/ropls.html) was used, which implements the PCA, PLS-DA and OPLS-DA approaches 55 . It includes R 2 and Q 2 goodness-of-fit and goodness-of-prediction statistics and permutation tests as well as scores, loadings, predictions, diagnostics and outlier graphics 55 . R 2 x describes the percentage of predictive and orthogonal variation in x that is explained by the full model; R 2 y describes the total sum of variation in y explained by the model; and Q 2 describes the predictive performance of the model calculated by full cross-validation.

LC-MS/MS analysis
Materials. IPA (Aldrich 1 g) as standard and indole-3-propionic-2,2-d 2 acid (0.05 g, C.D.N. Isotopes) were used as the IS solution. Acetonitrile, methanol and formic acid were of ULC-MS grade as supplied by Bio-Lab. Water with resistivity of 18.2 MΩ was obtained using the Direct 3-Q UV system (Millipore). A standard curve was built using a concentration range of IPA of 0.01-12 µg ml −1 , with a final IS concentration of 100 ng ml −1 .
Extract preparation. Plasma (30 µl) and IS (10 ul, 1 µg ml −1 ) were incubated for 10 min, and 500 µl of methanol was then added. The mixture was shaken at 10 °C for 30 min (ThermoMixer C, Eppendorf) and centrifuged at 21,000g for 10 min. Collected supernatant was evaporated by using a SpeedVac and then by a lyophilizer. Before LC-MC analysis, the obtained residue was resuspended in 100 µl of 20% aqueous methanol and then centrifuged twice at 21,000g for 5 min to remove insoluble material. The soluble part was placed in the insert of the LC-MS vial. Some samples were diluted 1:20 with 20% aqueous methanol.

16S rDNA amplicon sequencing
The caecum content was collected from mice and immediately frozen in liquid nitrogen. Samples were stored at −80 °C. Faecal DNA was extracted using a Fast DNA SPIN Kit (MP Biomedicals). Libraries were prepared using the 16S Metagenomic Sequencing Library Preparation protocol for the Illumina MiSeq System. In brief, PCR primers (16S Amplicon PCR Forward Primer, 5′-TCGTCGGCAGCGT CAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′;16S Amplicon PCR Reverse Primer, 5′-GTCTCGTGGGCTCGGAGATGTGTAT AAGAGACAGGACTACHVGGGTATCTAATCC-3′) directed at the V3/V4 region of bacterial rRNA genes were used to generate libraries, which were validated using an Agilent 2100 Bioanalyzer. Sequencing was performed with 250 bp paired-end reads on an Illumina MiSeq2500 at the NIHR Imperial Biomedical Research Centre (BRC) Genomics Facility.

16S rRNA amplicon sequencing data analysis
Only single-end R1 sequences were analysed. Data processing was carried out in R statistical environment version 3.5.2 according to the DADA2 pipeline 56 as follows. Primers were trimmed from demultiplexed sequences, amplicon sequence variants (ASVs) were generated using DADA2 package version 1.10.1 and taxonomic assignment was performed using SILVA rRNA database version 132 (https://www. arb-silva.de/). ASVs with no phylum assigned were excluded from analyses to yield a final number of n = 906 annotated ASVs. Moreover, α-and β-diversity analyses were carried out using phyloseq package version 1.26.1 (ref. 57 ). The α-diversity (Shannon Index) was calculated using raw counts 58 . Groups were compared using Mann-Whitney U statistical tests. For β-diversity analyses, data were normalized by log transformation with a pseudo-count of +1. Differential abundance of groups was compared with theDESeq2 package version 1.22.2 59 using the Wald test with Benjamini and Hochberg adjustments for multiple testing (P = 0.05). Whereas α-diversity is a measure of diversity or 'microbial variation' within a (single) sample, β-diversity is a measure of diversity across several samples. Piphillin was used to predict metagenomic alterations following the published methodology 60 . Briefly, metagenomic prediction analysis was carried out using the Piphillin web server. PUMAA 1.2 was used to annotate the predicted pathway abundances from the Piphillin output, and statistical analysis was carried out using STAMP 2.1.3.

RNA sequencing
DRG samples were collected from animals that had undergone 10 days of IPA (20 mg kg −1 per day) or PBS treatment naive or treatment followed by SNC. Sciatic DRGs were extracted 72 h after SNC or from naive animals (surgeries were performed as described above) and placed into RNAlater solution. The DRGs were crushed with an RNase-free micro-pestle, and the RNA was then immediately extracted using an RNAeasy Kit (Qiagen) according to the manufacturer's guidelines. Residual DNA contamination was removed by treating the spin column with 40 units of RNase-free DNase I (Qiagen) for 15 min at 23 °C before RNA elution. RNA concentrations and purity were verified for each sample following elution with the Agilent 2100 Bioanalyzer. RNA with integrity number (RIN) factors above 8.5 were used for library preparation. cDNA libraries for each sample were generated by the Imperial BRC Genomics Facility using the TruSeq Sample Preparation Kit A (Illumina) and were sequenced using Illumina HiSeq 4000 (PE 2 × 75 bp) sequencing. GO analysis was performed on differentially expressed genes using DAVID version 6.8 (http://david.abcc.ncifcrf.gov/). Differentially expressed genes were selected using a threshold of P < 0.05 and |1.5| < FC (fold change) or no FC cutoff. To identify IPA-dependent specific differentially expressed genes after SNC, we compared IPA-SNCvsPBS and PBS-SNCvsPBS groups and selected the uniquely upregulated or downregulated genes in the IPA-SNCvsPBS group. The heat map was created using GraphPad Prism 8.0

Sciatic nerve regeneration
Twenty-four or 72 h after the surgery, sciatic nerves were dissected and post-fixed in 4% paraformaldehyde (PFA), incubated at 4 °C for 1 h and transferred into 30% sucrose for at least 3 days. Subsequently, the tissue was embedded and frozen in Tissue-Tek optimum cutting temperature solution and was maintained at −80 °C until being cut into 11-µm sagittal sections. Tissue sections were immunostained for SCG-10 (1:1,000, rabbit, Novus) a marker for regenerating axons. The crush site was identified by deformation of the nerve and disruption of axons coinciding with the highest SCG-10 intensity. The SCG-10 intensity was measured at 500-µm intervals along the length of the nerve distal to the SNC site. The intensity was normalized to the SCG-10 intensity before the crush site and was plotted as FC. Four to six sections per animal were analysed and imaged using an HW1 Zeiss Axio Observer with a Hamamatsu Flash 4.0 fast camera at ×10 magnification.

DRG cell culture
Glass coverslips were coated with 0.1 mg ml −1 poly(d-lysine), washed and coated with 2 µg ml −1 mouse laminin (Millipore) for 1-2 h each before the start of the experiment. Sciatic DRGs from adult animals were dissected and collected in Hanks balanced salt solution (HBSS) on ice. The DRGs were then transferred into a digest solution (5 mg ml −1 Dispase II (Sigma), 2.5 mg ml −1 collagenase type II (Worthington)) and incubated in a 37 °C water bath for 45 min, with occasional shaking for 30 s. Thereafter, the DRGs were washed and manually dissociated with a 1-ml pipette in medium containing 10% heat-inactivated FBS (Invitrogen) and 1× B27 (Invitrogen) in F12:DMEM (Invitrogen). Pipetting was continued until the DRGs were fully dissociated and no clumps could be observed. Next, the cell suspension was spun down at 1,000 r.p.m. for 4 min and was resuspended in culture medium containing 1× B27 and penicillin/streptomycin in F12:DMEM. Next, 3,500 cells were plated on each coverslip (coated with poly(d-lysine) and laminin) and were maintained in a humidified culture chamber with 5% CO 2 at 37 °C for 12 h before being fixed with 4% PFA and immunostained.
For IFNγ-dependent ex vivo DRG outgrowth, animals were injected i.p. with 10 µg IFNγ per mouse 48 h before DRG cell culturing.

Immunocytochemistry
Plated cells were fixed by incubation with cold 4% PFA for 15 min and were then blocked and permeabilized for 1 h with 0.3% TX100 in PBS containing 2% BSA. The primary antibody staining was performed using anti-βIII-tubulin (1:1,000, mouse, Promega) in 0.1% TX100 in PBS containing 2% BSA with overnight incubation at room temperature. The goat secondary antibody (Alexa Fluor) was diluted in 0.1% TX100 in PBS containing 2% BSA, and the cells were incubated for 1 h. All cells were counterstained with DAPI.

FACS
Cells were isolated from spleens or lymph nodes by mashing using a 70-µm cell strainer and a pestle and were transferred into media (RPMI 1640 (Thermo Fisher Scientific) + 2% FCS + 1× penicillin/streptomycin + 1× HEPES (Thermo Fisher Scientific)). Spleen cells were centrifuged at 1,200 r.p.m. for 6 min and resuspended in 1 ml of Hybrid Max red blood cell lysis buffer (Sigma) for 5 min. Next, 9 ml of medium was added, and cells were centrifuged at 1,200 r.p.m. for 6 min. Spleen cells or lymph node cells were resuspended in medium and plated at 1 × 10 6 to 2 × 10 6 cells per well in a 96-well plate. The plate was centrifuged at 2,000 r.p.m. for 1 min, and 50 µl of blocking stain was added, which contained 1:50 dilution rat serum (Sigma) and a 1:50 dilution of TruStain FcX (anti-mouse CD16/32, Biolegend) in FACS buffer (PBS supplemented with 5% FCS, 2 mM EDTA and 0.09% sodium azide). Cells were blocked for 30 min at 4 °C and were subsequently washed three times with 140 µl of FACS buffer. Antibodies were added, and the cells were incubated for 30 min in the dark. Next, the cells were washed three more times and were then examined using a BD LSR Fortessa X-20 (BD Biosciences, 4 Laser Flow Cytometer). FlowJo (version 10.7.1) was used for data analysis.

Microscopy
Photomicrographs were taken with a Nikon Eclipse TE2000 microscope with an OptiMOS scientific complementary metal-oxide semiconductors (cMOS) camera using a ×10-or ×20-resolution Zeiss Axio Observer with a Hamamatsu Flash 4.0 fast camera at ×10 or ×20 magnification. Confocal images were taken with a Leica TCS SP8 II confocal microscope at ×40 or ×60 magnification and were processed with LAS-AM software (Leica).

Image analysis for immunohistochemistry and immunocytochemistry
Image analysis was conducted using ImageJ (Fiji) software. All analysis was performed by the same experimenter, who was blinded to the experimental groups. DRG images were taken using a Nikon Eclipse TE2000 microscope with an OptiMOS scMOS camera at either ×10 or ×20 magnification. Images were analysed by counting the number of cells per area and measuring the intensity against the background or calculating the percentage of cells with positive staining. For markers expressed in all cells, only the intensity was assessed.
For neurite length analysis, between 15 and 20 images were taken per coverslip and were analysed using ImageJ (ImageJ-win64) NeuronJ plugin software. All analyses were performed with blinding. Approximately 45-60 cells were analysed per animal and condition. ImageJ (ImageJ-win64) Macromanager 1.4 software was used for all image acquisition.

Statistics and reproducibility
Results are graphed as mean ± s.e.m. Statistical analysis was carried out using GraphPad Prism 8 software. Normally distributed data were evaluated using a two-tailed unpaired Student's t-test or one-way ANOVA when experiments contained more than two groups. The Dunnett multiple-comparisons test was applied when appropriate. Two-way ANOVA with Tukey's or Sidak's multiple-comparisons test was applied when two independent variables on one dependent variable were assessed. The threshold level of significance was set at P < 0.05. All data analysis was performed with blinding to the experimental group. All data were obtained over two to three independent experiments. The sample size was based on similar, previously established experimental designs or was calculated using the Animal Experimentation Ethics Committee power calculator (http://www.lasec.cuhk.edu. hk/sample-size-calculation.html) to estimate the number of replicates required, for a difference of 1.5 and 80-90% power, assuming a 5% significance level.

Replication
All attempts at replication were successful.

Blinding
Nerve regeneration was assessed with blinding by two independent scientists. The Von Frey and Hargreaves tests were assessed using two experimenters, one of whom was blinded to the experimental groups.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
All RNA sequencing data are available from the NCBI GEO database under accession number GSE161342. Source data are provided with this paper.  Extended Data Fig. 6 | RNA sequencing from DRG following IPA treatment and injury and increased number of neutrophils in the DRG following IPA treatment. a-b. Shown are heatmaps of differentially regulated genes (FPKMs, P < 0.05) following 10 days IPA vs PBS treatment as, preceding (a) and 3 days following SN crush (b). Green: downregulated, red: upregulated. c. Heatmap showing the logarithmic expression fold change (logFC (P < 0.05)) of all genes belonging to "Neutrophil-Endothelial interaction" or "BP Neutrophil chemotaxis" GO class for the comparisons IPA-SNCvsPBS, PBS-SNCvsPBS and IPAvsPB. d. Representative image of a DRG stained for Ly6G (red), ßIIITubulin (ßIIITub, green) and DAPI (blue). Scale bar: 100 µm. e. Quantitative analysis of neutrophils, shown as number of neutrophils per 1 mm 2 (N = 4 biologically independent animals per group, One-way Anova with Holm-Sidak multiple comparison test, if not indicated otherwise p-value compares to PBS group, data are presented as mean values +/ − SEM). f. Representative images of Ly6G+ cells in DRG tissue. Scale bar: 50 µm. g. Images of the sciatic nerve crush site, immunostained for Ly6G (red) after IPA and PBS treatment. Asterisk indicates the crush site. Scale bar: 250 µm. Magnified image scale bar: 50 µm. h. Quantitative analysis showing the number of neutrophils 1000 µm proximal and distal to the crush site (N = 4 biologically independent animals per group, data are presented as mean values +/ − SEM). i. FACS analysis of the nerve crush site following 10 days of IPA treatment and 3 days SNC. j+k. Bar graphs showing Ly6G+ cell counts (g) and percentage of Ly6G+ cells of the total CD45 + cells (h) in nerve crush site 3 days following SNC (N = 3 biologically independent animals per group, data are presented as mean values +/ − SEM). Fig. 7 | FACS gating strategies. a. Gating strategy for the quantification of Ly6G+hi neutrophils (as % of CD45+ and total cell counts) from the nerve crush site following IPA or PBS treatment (corresponding to Extended Data Fig. 6i-k). b. Gating strategy for the verification of neutrophil depletion following αLy6G or αIgG2A treatment (corresponding to Extended Data Fig. 8h, i). Gr-1+ neutrophils were quantified from the spleen as percentage of CD45+.