Abstract

   Type 1 diabetes (T1D) is linked to an altered gut microbiota
   characterized by reduced short-chain fatty acid (SCFA) production. Oral
   delivery of a SCFA-yielding biotherapy in adults with T1D was followed
   by increased SCFAs, altered gut microbiota and immunoregulation, as
   well as delaying diabetes in preclinical models. Here, we show that
   SCFA-biotherapy in humans is accompanied by remodeling of the gut
   proteome and mucosal immune homeostasis. Metabolomics showed arginine,
   glutamate, nucleotide and tryptophan metabolism were enriched following
   the SCFA-biotherapy, and found metabolites that correlated with
   glycemic control. Fecal microbiota transfer demonstrated that the
   microbiota of SCFA-responders delayed diabetes progression in humanized
   gnotobiotic mice. The protected mice increased similar metabolite
   pathways to the humans including producing aryl-hydrocarbon receptor
   ligands and reducing inflammatory mucosal immunity and increasing IgA
   production in the gut. These data demonstrate that a potent SCFA
   immunomodulator promotes multiple beneficial pathways and supports
   targeting the microbiota as an approach against T1D. Trial
   registration: Australia New Zealand Clinical Trials Registry
   ACTRN12618001391268.

   Subject terms: Microbiome, Autoimmunity, Type 1 diabetes
     __________________________________________________________________

   In a previous clinical trial, the authors reported that SCFA-yielding
   biotherapy in adults with type 1 diabetes remodels the gut proteome and
   metabolome. Here, the show that colonization of the post-therapy
   microbiome into mice delayed diabetes and increased aryl-hydrocarbon
   receptor ligand and IgA production in the gut.

Introduction

   The gut microbiota is a key drivers of intestinal homeostasis and
   regulators of the developing immune system^[64]1. Dysbiosis of the gut
   microbiota has been implicated as a possible factor contributing to the
   development of autoimmune diseases such as type 1 diabetes (T1D)^[65]2.
   T1D is caused by a T-cell driven autoimmune attack on the
   insulin-producing beta cells of the pancreatic islets and is commonly
   diagnosed in childhood. Multiple studies have reported alterations in
   the gut microbiota composition preceding T1D onset, with increases in
   Bacteroides species, decreases in Firmicutes and decreased beta
   diversity all reported^[66]3. While the taxonomic changes observed are
   not always consistent between studies, a common functional deficiency
   has been related to T1D at various disease stages. A reduction in SCFA,
   acetate-, propionate- and butyrate- producing bacteria was associated
   with islet autoimmune responses in mice, newly diagnosed children, and
   in adults with longstanding T1D^[67]4–[68]7. Thus, a functional defect
   in microbial SCFA production, rather than a deficit in specific
   species, may be linked to the progression of islet autoimmunity and
   T1D^[69]8.

   SCFAs are well known for their wide-ranging beneficial effects on the
   host, acting as an energy source, signaling molecules and as epigenetic
   regulators^[70]9. Butyrate is a major energy source for the colonic
   epithelium, improving intestinal homeostasis, stimulating mucous
   production, and promoting barrier function through tight-junction
   formation^[71]9. In immune cells, butyrate promotes regulatory T cell
   (Treg) differentiation and suppresses inflammatory cytokine production
   by antigen presenting cells, while acetate modulates B-cell activation
   and IgA production^[72]4,[73]7,[74]10,[75]11. SCFAs also improve islet
   beta-cell survival in response to stress^[76]12,[77]13. In addition to
   SCFAs, other microbially derived metabolites including arginine,
   tryptophan metabolites, glutamic acid, succinic acid, ornithine and
   citrulline have been associated with altered risk of future progression
   of T1D^[78]14–[79]17. We hypothesize that a high-SCFA yielding
   intervention will remodel the gut microbiota, restoring downstream
   production of multiple microbial metabolites required to reprogram the
   progression of islet autoimmunity.

   Previously, we demonstrated that a SCFA-biotherapy comprising of a
   polysaccharide biomolecule modified to deliver acetate and butyrate
   (HAMSAB) protects from diabetes development in the non-obese diabetic
   (NOD) mouse model of T1D^[80]4. The mechanism of action was due to
   reduced activation of islet-specific T cells, expanded Tregs and a
   reduction in hyperactive B cells^[81]4. In addition, this potent
   immunomodulator has been shown to have therapeutic effects in
   hypertension^[82]18, which is an indicator for the effectiveness of
   SCFAs in treating serious or life-threatening conditions like T1D. In a
   recent pilot study, adults with long-standing T1D given SCFA-biotherapy
   for 6 weeks had increased fecal and plasma SCFAs and an altered gut
   microbiota^[83]19. This correlated with a shift in circulating immune
   response, characterized by decreased B-cell activation and elevated
   regulatory TIGIT and CTLA4 expression in T cell subsets. Though not
   designed to assess beta cell preservation or T1D delay, the study’s
   findings suggest that the effects of SCFA-biotherapy on immune
   regulation observed in preclinical models apply to humans.

   The aim of this study was to determine if SCFA-biotherapy altered
   mucosal immunity, gut barrier function and metabolite production,
   potentially reducing T1D onset. Using fecal proteomics, we showcase
   remodeled gut barrier proteins and innate immunity following the SCFA
   intervention. We identified metabolic shifts in arginine, nucleotide,
   and tryptophan metabolites, and then correlated these with glycemic
   control. We established a causal connection between SCFA-modified gut
   microbiota and its role in preventing or delaying T1D by transferring
   fecal microbiota from SCFA responder and non-responder human
   participants into germ-free NOD mice. We showed that the microbiota
   from SCFA-responders increased butyrate production and delayed diabetes
   progression after transfer into NOD mice, which was associated with a
   distinct metabolite profile, aryl hydrocarbon receptor ligand
   production and remodeled mucosal immunity. With broad effects on both
   gut barrier, immunity and metabolism, restoration of SCFA production is
   a promising approach for interventions designed to halt progression of
   T1D.

Results

Fecal human proteome remodeling following SCFA-yielding biotherapy

   In NOD mice, administration of an acetate and butyrate yielding
   biotherapy improved gut homeostasis via increasing tight-junction and
   IL22 gene expression^[84]4. We hypothesized that increasing SCFA
   delivery would also impact on mucosal immunity and gut barrier function
   in humans. To investigate this, we used stool samples collected from
   adults with long-standing T1D during a 6-week intervention of a
   SCFA-biotherapy, with study visits at week 0 (baseline), week 6 (end of
   intervention) and week 12 (follow-up), as previously described^[85]19.
   The study was a single-armed pilot trial and enrolled 21 participants,
   of which 18 completed the trial with stool and plasma samples used in
   the current study (Fig. [86]1A). The characteristics of the study
   participants are described in Supplementary Data [87]1.

Fig. 1. Gut barrier protein remodeling revealed through fecal proteomics
after SCFA biotherapy.

   [88]Fig. 1
   [89]Open in a new tab

   A Schematic overview of the trial. Created in BioRender.
   Hamilton-Williams, E. (2025) [90]https://BioRender.com/h13m098. B
   Multivariate PLS-DA of human stool proteins. Significance determined by
   two-sided PERMANOVA. Pairwise comparisons were adjusted for multiple
   correction by false discovery rate. Ellipses show the 95% confidence
   interval. C, D PLS-DA components 1 and 2, top 20 loadings. Orange:
   baseline, blue: week 6, red: follow-up. E Top-40 significantly enriched
   Gene Ontology biological processes (FDR < 0.05) enriched using proteins
   (p < 0.05) that increased from baseline to 6 weeks. F KEGG pathway
   enrichment of proteins which increased from baseline to six weeks or G
   decreased from baseline to six weeks. Significance determined by linear
   mixed effects models using a Likelihood Ratio Test (p < 0.05) and
   Benjamini-Hochberg correction for multiple testing. Enrichments were
   performed using proteins with p < 0.05. The cut-off for edges is 40%
   shared proteins. Thickness of edges indicate degree of protein overlap.
   See also Supplementary Fig. [91]1 and Supplementary Data [92]2-[93]4.

   To investigate gut barrier homeostasis in these individuals, we
   performed untargeted proteomic analysis of soluble stool proteins,
   which are enriched for human proteins derived from the gut epithelium
   and exocrine pancreas^[94]20. Searching the detected peptides against
   the human proteome database identified 738 human proteins after
   filtering. Supervised clustering using partial least-squares
   discriminant-analyses (PLS-DA) indicated the host fecal proteome was
   unique at all three timepoints and this was confirmed by PERMANOVA
   (p = 0.001), with the follow-up timepoint not returning to baseline
   after the intervention ceased (Fig. [95]1B-D). Univariate linear mixed
   model analysis found 159 proteins that differed between the timepoints
   (p < 0.05), with 53 of these having a false discovery rate (FDR) < 0.05
   (Supplementary Data [96]2, Supplementary Fig. [97]2).

   Gene Ontology pathway enrichment analysis of proteins that increased
   following SCFA-biotherapy found 201 enriched biological processes at 6
   weeks (FDR < 0.05, Fig. [98]1E, Supplementary Data [99]3). The most
   enriched process was cell redox homeostasis, which included
   antioxidants peroxiredoxin-6 (PRDX6), thioredoxin (TXN), and
   thioredoxin reductase 1 (TXNRD1), which protect from oxidative
   stress^[100]21. Concordant with this, proteins associated with response
   to hypoxia and response to decreased oxygen levels were enriched,
   suggesting a lower oxygen environment. Actin filament pathways, fiber
   organization, and epithelial cell differentiation were increased,
   driven by epithelial-cell junction proteins CAPZB, CAPZA1, and radixin
   (RDX). Multiple innate immune processes were enriched related to
   neutrophils and myeloid cells including anti-inflammatory protease
   progranulin (GRN), anti-microbial defense protein NOX2 and proteins
   involved in cell adhesion, migration and angiogenesis such as
   grancalcin (GCA), integrin subunit alpha M (ITGAM or CD11b), and matrix
   metallopeptidase-9 (MMP9) (Supplementary Fig. [101]1, Supplementary
   Data [102]3). At follow-up, epithelial and neutrophil and myeloid cell
   pathways remained enriched demonstrating persistent effects after the
   end of the intervention (Supplementary Data [103]3).

   Pathway enrichment using KEGG found 19 pathways were significantly
   enriched at 6 weeks (FDR < 0.05, Fig. [104]1F, Supplementary
   Data [105]4). These included innate immune pathways, indicating a
   strong involvement of neutrophils and phagosomes. Tight junction
   pathways driven by TUB1A, RDX, and SLC9A3R1 increased (Supplementary
   Fig. [106]1), as were protein processing in the endoplasmic reticulum
   including heat shock proteins involved in suppressing aggregation of
   mis-folded proteins (CRYAB, HSP90B1, and PRKCSH). Enrichment of
   alanine, aspartate, and glutamate metabolism and arginine biosynthesis
   included arginosuccinate synthase-1 (ASS1), CAD protein (CAD) and
   4-aminobutyrate aminotransferase (ABAT), nitric oxide synthase (NOS2)
   and aldehyde dehydrogenase 2 (ALDH2), which all increased after
   SCFA-biotherapy (Supplementary Fig. [107]1). At follow-up, lysosome and
   phagosome pathways were enriched suggesting durable innate immune
   responses (Supplementary Data [108]4).

   KEGG pathways that decreased after SCFA biotherapy included ferroptosis
   and necroptosis cell death pathways (Fig. [109]1G, Supplementary
   Data [110]4). These were driven by iron storage proteins, ferritin
   light chain and ferritin heavy chain 1, and the voltage-dependent
   anion-selective channel protein, a prop-apoptotic protein
   (Supplementary Fig. [111]1). Necroptosis is an inflammatory form of
   cell death often resulting from pathogen-induced cell damage^[112]22.
   Ferroptosis is an iron-associated form of cell death linked to a
   failure of antioxidant defenses and is associated with dysfunction of
   the intestinal epithelium and gut disease^[113]22. Similar pathways
   persisted at follow-up with necroptosis remaining downregulated
   (Supplementary Data [114]4). Together, these data suggest that a
   neutrophil-mediated tissue-repair process may be activated following
   SCFA biotherapy, accompanied by an increase in tight junction proteins
   and a reduction in cell-death pathways.

Metabolite production in the gut is significantly altered following SCFA
biotherapy

   As metabolites produced by the gut microbiota are likely mediators of
   any effects of SCFA biotherapy on the gut epithelium, we used
   untargeted metabolomics to broadly profile fecal metabolites, including
   metabolites produced by the microbiota and the host. LC/MS analysis
   detected 603 putatively annotated metabolites. Supervised multivariate
   analysis showed that the fecal metabolome at 6 weeks was distinct from
   the other two timepoints (PERMANOVA p = 0.006, Fig. [115]2A,
   Supplementary Fig. [116]2A). Univariate analysis found 101 metabolites
   with p < 0.05, of which eight had an FDR < 0.05 (Fig. [117]2B,
   Supplementary Fig. [118]2B, Supplementary Data [119]5). MetOrigin
   pathway analysis identified an enrichment of multiple pathways
   including alanine, aspartate and glutamate metabolism, butanoate
   metabolism, arginine biosynthesis, TCA cycle, and purine metabolism at
   both 6 weeks and follow-up, with the majority of pathways shared by
   both host and microbiota (Fig. [120]2C, D, Supplementary Data [121]6).
   Pyrimidine metabolism, driven in part by microbiota-specific pathways,
   was enriched at follow-up (Fig. [122]2C, D, Supplementary Data [123]6).
   Within these pathways, increases in succinic acid (succinate) and
   butyric acid (butyrate) at both 6 and 12 weeks contributed to the
   butanoate pathway enrichment (Supplementary Fig. [124]2C). Butyrate is
   a major source of energy for colonocytes through the TCA cycle (which
   includes succinate)^[125]23. Arginine pathway enrichment was driven by
   increased 1-pyrroline-5-carboxylic acid at 6 weeks and increased
   creatine at 12 weeks and decreases in ornithine, glutamine, and
   2-oxoarginine Supplementary Fig. [126]2C, Supplementary Data [127]6).
   Alanine, aspartate, and glutamate pathway enrichment was driven by
   decreases in alanine, oxoglutaric acid, and glutamine, but increased
   succinic acid (Supplementary Fig. [128]2C, Supplementary Data [129]6).
   Purine and pyrimidine nucleotide metabolism pathways included
   deoxyinosine, deoxyuridine, and pseudouridine, which all tended to
   increase at 6 weeks, and significantly dropped at 12 weeks
   (Supplementary Fig. [130]2B, C, Supplementary Data [131]6).
   Deoxyinosine and deoxyuridine, along with a tripeptide containing a
   glutamic acid and two valines (Glu/Val/Val),
   4-hydroxy-2-oxo-heptanedoic acid, and gamma-glutamylalanine negatively
   correlated with HbA1c and positively correlated with estimated
   C-peptide (Fig. [132]2E).

Fig. 2. Gut metabolite production undergoes significant changes after SCFA
biotherapy.

   [133]Fig. 2
   [134]Open in a new tab

   A Multivariate PLS-DA of human stool metabolites. Significance
   determined by two-sided PERMANOVA. Pairwise comparisons were adjusted
   for multiple corrections by false discovery rate. Ellipses show a 95%
   confidence interval. B Abundance heatmap of stool metabolites with an
   FDR < 0.1 determined by linear mixed effects model using a Likelihood
   Ratio Test (p < 0.05) and Benjamini-Hochberg correction for multiple
   testing. Metabolites in bold are level 1 (identified using reference
   standards). C Significant KEGG pathways enriched from baseline to six
   weeks and D baseline to follow-up. Enrichments were performed in
   MetOrigin using metabolites with p < 0.05. Differences were determined
   using paired two-tailed t-test of individual metabolites with a cut-off
   value of p < 0.05. Green bars: microbiota pathways, black bars shared
   host and microbiota (co-metabolism) pathways. E Repeated measures
   correlations of metabolites with FDR < 0.1 with glycemic variables. *
   indicates FDR corrected p-value > 0.01 < 0.05 and · indicates
   p-value > 0.05 < 0.1 F Pathway schematic showing key human stool
   proteins, stool metabolites, and metabolic pathways altered after the
   intervention. Solid arrow indicates changes at 6-weeks, dashes arrow
   indicates changes at follow-up. TCA cycle: The citric acid cycle, NO:
   nitric oxide. Created in BioRender. Hamilton-Williams, E. (2025)
   [135]https://BioRender.com/l00t262. See also Supplementary Fig. [136]2
   and Supplementary Data [137]2, [138]5, and [139]6.

   A combined analysis of the stool metabolites, pathways and stool
   proteins supports the hypothesis that the SCFA biotherapy is driving
   the TCA cycle and alanine, aspartate, and glutamate metabolism leading
   to glutamate and glutamine being converted into purine and pyrimidine
   nucleotides, as well as driving the urea cycle and arginine metabolism
   (pathway schematic shown in Fig. [140]2F). Most changes from baseline
   to 6 weeks were precursor metabolites involved in arginine biosynthesis
   such as the stool metabolites N-acetyl-glutamic acid, glutamine and
   glutamate. Whereas changes from baseline to follow-up were metabolites
   involved in arginine utilization such as the stool metabolites
   creatine, creatinine and 2-oxoarginine (Fig. [141]2F).

Plasma metabolomics reveals systemic metabolic effects following
SCFA-biotherapy

   To further understand how the 6-week SCFA-biotherapy in adults with
   T1D, and the associated changes in the microbiome and metabolome
   influenced the host systemically, we investigated circulating
   metabolites in plasma. Untargeted LC/MS analysis of plasma identified
   619 putatively annotated metabolites. Multivariate PLS-DA showed the
   greatest effect on plasma metabolites at follow-up, and this was
   significant by PERMANOVA (p = 0.022, Fig. [142]3A, B). Univariate
   analysis found 92 metabolites had a p < 0.05 using a linear mixed model
   (Supplementary Data [143]7). Succinic acid was increased at follow-up
   compared with 6 weeks and 2-oxoarginine decreased at week 12, similar
   to the stool metabolome (Supplementary Data [144]7). An exploratory
   enrichment analysis found tryptophan metabolism and phenylalanine,
   tyrosine and tryptophan biosynthesis were the most impacted pathways at
   12 weeks (Fig. [145]3C, Supplementary Data [146]8), which was
   attributed to tryptophan, kynurenine, and indolepyruvate (Supplementary
   Data [147]8). Valine, leucine, and isoleucine biosynthesis and
   degradation were also enriched. These data suggest that there is a
   sustained systemic impact on metabolic pathways following SCFA
   biotherapy.

Fig. 3. Plasma metabolites are significantly altered at follow-up.

   [148]Fig. 3
   [149]Open in a new tab

   A PLS-DA analysis of plasma metabolites and B loading plot for
   component 1 of the PLS-DA. Significance determined by two-sided
   PERMANOVA. Pairwise comparisons were adjusted for multiple correction
   by false discovery rate. Ellipses indicate 95% confidence intervals.
   Orange: baseline, blue: week 6, red: follow-up. Metabolites in bold are
   level 1 (identified using reference standards). C MetaboAnalyst pathway
   enrichment analysis of plasma metabolites that changed from 6-weeks to
   follow-up and baseline to follow-up. Enrichment used metabolites with
   two-sided linear mixed model analysis p < 0.05. Red dashed line
   indicates pathway significance. Color represents the significance. Dot
   size represents impact. P values are provided. See also Supplementary
   Data [150]7 and [151]8.

Human to mouse microbiota transfer model demonstrates disease-modifying
effects of the SCFA-biotherapy modified microbiota

   We next used a human-to-mouse fecal microbiota transfer model to
   investigate whether the functional changes in the human microbiota
   present after the SCFA-biotherapy could modify disease progression
   (Fig. [152]4A). Given HAMSAB significantly increased the concentrations
   of acetate, butyrate and propionate in feces and circulation in both
   mice^[153]4 and humans^[154]19, we classified participants from the
   trial based on their increase in stool SCFAs at 6-weeks, with a fold
   change ≥2 termed ‘responders’ and <2 as “non-responders”
   (Fig. [155]4B). We selected n = 2 SCFA non-responders (donors A and B)
   and n = 2 responders (donors C and D) that also had a high live
   bacteria count. While responder and non-responder subjects similarly
   altered their alpha and beta diversity following the biotherapy
   Supplementary Fig. [156]3, the responder individuals tended to have
   higher alpha diversity. Resuspensions of individual human donor stool
   samples were gavaged into female germ-free NOD mice prior to mating and
   the female offspring was studied for microbiota colonization,
   metabolite levels, and diabetes progression (Fig. [157]4A). A control
   group of dams was gavaged with sterile saline (PBS) and their offspring
   remained germ-free.

Fig. 4. Human-to-mouse microbiota transplantation from responders delays
diabetes in NOD mice.

   [158]Fig. 4
   [159]Open in a new tab

   A Experimental design. Female germ-free NOD mice were colonized with a
   human stool suspension or remained uncolonized (PBS controls) by oral
   gavage on days 0 and 7, then mated one week later. Female offspring
   were followed for diabetes progression and fecal samples collected for
   analysis. Created in BioRender. Hamilton-Williams, E. (2025)
   [160]https://BioRender.com/u03r698. B Donor samples, represented by
   blue dots (non-responder donors) and red dots (responder donors), from
   four participants were selected from the n = 18 completed SCFA
   biotherapy trial participants based on fold-change in fecal butyrate at
   6 weeks from baseline. One outlier individual is excluded from the
   graph with fold-changes of 73.2, 123.8, and 299.5. The dashed line
   represents fold-change of 2 cut-off for defining responder status. C
   Fecal butyrate in the female offspring of the mice (7–9 weeks of age)
   colonized with responder (red) or non-responder (blue) human donor
   stool samples (baseline and 6-week trial timepoints), or control
   germ-free mice. The line represents median value. Significance based on
   one-way ANOVA with Tukey’s correction. Survival curve analysis
   (log-rank Mantel-Cox) test in germ-free NOD mice colonized with
   non-responder vs responder stool samples collected from donors at
   6-weeks (D) or baseline (E) or control germ-free mice which received a
   sterile PBS gavage. Diabetes data was pooled from two independent
   experiments. F Pancreas sections stained by immunofluorescence for
   insulin (green) and glucagon (magenta), with DAPI nuclear stain (blue)
   from female colonized mice at 10 weeks of age (n = 7 responder and
   n = 9 non-responder mice, from one experiment). Representative image is
   shown at 20× magnification. See also Supplementary Fig. [161]3–[162]6.

   All of the mice remained on a standard diet throughout the experiment.
   Interestingly, fecal butyrate in the offspring was the only SCFA
   significantly higher from the mice colonized from the post-intervention
   responder samples (6-week donor samples, p = 0.0282) compared to the
   non-responder colonized mice and higher than both groups colonized with
   the baseline donor samples (Fig. [163]4C, Supplementary Fig. [164]4,
   p = 0.0061).

   The mice colonized with the post-SCFA-biotherapy microbiota from
   responders at 6 weeks showed a notable delay in diabetes progression
   compared to both the non-responder group at the same time point
   (p = 0.0198) and the PBS control germ-free group (p = 0.0087)
   (Fig. [165]4D). In contrast, mice colonized with the baseline samples
   from either the responders or non-responders failed to significantly
   delay diabetes in colonized mice compared to the PBS control group
   (Fig. [166]4E). A second cohort of germ-free female NOD mice were
   similarly colonized and analyzed at 10-weeks of age for their immune
   cell and pancreas phenotype. Histology and immunofluorescence staining
   trended to reduced mononuclear cell infiltration and insulin content in
   islets from the responder compared to the non-responder colonized mice
   (Fig. [167]4F, Supplementary Fig. [168]5). A lower proportion of BDC2.5
   islet-specific CD4^+ T cells identified by tetramer staining were found
   in the pancreatic lymph nodes of the responder mice colonized with the
   post-biotherapy microbiota (p = 0.0448, Supplementary Fig. [169]6).
   Additionally, the responder mice had increased Foxp3^+CD8^+ regulatory
   T cells in the spleen (p = 0.053) and pancreatic lymph nodes
   (p = 0.0099). The data suggest that the microbial changes caused by
   SCFA intervention in humans still impact the development of
   autoimmunity in mice after adoptive transfer, with the protective
   effects primarily associated with butyrate response.

Diabetes-protected responder colonized mice cluster by fecal metabolites, not
microbiota composition

   Colonization of the fecal microbiota in both the gavaged mothers and
   the female offspring was investigated using 16S rRNA sequencing.
   Similar to previous studies^[170]24, alpha diversity in the colonized
   mice samples was lower than in the human donor samples (Supplementary
   Fig. [171]7). However, multivariate PLS-DA showed that the human donor
   samples clustered with their respective colonized mouse samples
   (Supplementary Fig. [172]7). While there were differences in abundance
   of taxa between the donor human samples and the mice, there was
   colonization with 50.1% ± 3.6% (mean ± SD) of donor-derived species
   into the dams. High transmission occurred from the mothers to the
   offspring (mean 44.1% ± 8.3% of donor species present at 5 weeks of
   age), and colonization was relatively stable across time (mean
   44.8% ± 9.9% of donor species present at 11 weeks of age),
   (Supplementary Fig. [173]8). Unsupervised PCA multivariate analysis
   demonstrated that the microbiota composition of the mice colonized from
   each of the four donors was distinct (Fig. [174]5A, Supplementary
   Fig. [175]9A, B, PERMANOVA p < 0.001). While there was clustering by
   timepoint (baseline or 6-week donor samples), the donor effect was
   dominant, with 83 taxa significantly differing between the four donors
   (FDR < 0.05, Supplementary Data [176]9). Notably, the “responder” and
   “non-responder” donors did not cluster together, with donor D more
   dissimilar than the other three groups. We still found that 42 taxa
   significantly differed between the responder and non-responder groups
   with genera including Parasutterella, Lachnospiraceae, Alistipes, and
   Blautia higher in responders while Enterococcus, Turicibacter, and
   Butyricimonas were higher in non-responder colonized mice
   (Supplementary Data [177]10).

Fig. 5. Fecal metabolites, not gut microbiota composition, drive clustering
of recipient mice in human-to-mouse microbiota transplantation study.

   [178]Fig. 5
   [179]Open in a new tab

   A Sparse-PCA of 16S rRNA amplicon sequencing of feces collected from
   female NOD offspring of human-microbiota colonized mice at 5 weeks of
   age. Mice colonized with 6-week donor samples were shown. B PCA of
   fecal metabolites from female NOD mice pups collected at 7–9 weeks of
   age. Mice colonized with 6-week donor samples only were analyzed for
   metabolites. Significance determined by two-sided PERMANOVA. C Pathway
   enrichment analysis of metabolites from mice colonized with responder
   versus non-responder donors using MetOrigin. Differences were
   determined using a paired two-sided t-test of individual metabolites
   with a cut-off value of p < 0.05. Green bars— microbiome pathways,
   black bars—co-metabolism (shared between host and microbiota) pathways.
   D Abundance (log2 normalized intensity) of key metabolites from
   pathways identified in the enrichment analysis. Metabolites marked with
   # are microbiota only pathways determined from the origin-based MPEA
   analyses. 4-(2-Amino-3-hydroxyphenyl)-2,4-dioxobutanoic acid is the
   complete name for abbreviated metabolite in the tryptophan pathways.
   Box and whisker plots show median and inter-quartile range, with lines
   representing the minimum and maximum values. Schematic of E purine
   metabolism pathway and F pyrimidine metabolism pathway. Metabolites in
   bold are level 1 (identified using reference standards). Metabolites
   shown in red were higher in responder mice. Schematics created in
   BioRender. Hamilton-Williams, E. (2025)
   [180]https://BioRender.com/w72e215 and
   [181]https://BioRender.com/s39p554. See also Supplementary
   Fig. [182]7–[183]9 and Supplementary Data [184]9−[185]12.

   We next profiled the fecal metabolome of the colonized offspring
   (6-week sample colonized groups only), identifying 527 metabolites
   after filtering. In contrast to the microbiota composition, PCA
   multivariate analysis showed that the two responder groups had a
   similar metabolite profile distinct to the non-responders
   (Fig. [186]5B, Supplementary Fig. [187]9C). PERMANOVA analysis
   confirmed that the metabolites significantly differed between the
   donors (p < 0.001), with 302 individual metabolites differing between
   the responders and non-responders (FDR < 0.05, Supplementary
   Data [188]11). Pathway enrichment found that purine, pyrimidine,
   arginine, and glutamate metabolism were the most enriched pathways in
   the responder compared to the non-responder group (Fig. [189]5C,
   Supplementary Data [190]12). Multiple purines and pyrimidine pathway
   constituents were higher in responders, as were arginine precursor
   metabolites glutamate and oxoglutaric acid, while arginine itself and
   ornithine were lower (Fig. [191]5D, E). In addition, tryptophan
   metabolism was altered in the responder mice (Fig. [192]5D), driven by
   a highly significant elevation of quinic acid, a microbiota-only
   pathway metabolite, in responder mice (FDR = 2.18E-08). These
   metabolite changes and pathways enriched in the responder group showed
   a high similarity to the enriched pathways observed after SCFA
   biotherapy in the human trial participants (Fig. [193]2C, D).

Responder colonized mice have reduced mucosal B cell gene networks

   We next investigated mucosal immunity and the gut barrier in the mice
   colonized with the responder versus non-responder donor microbiota via
   transcriptomics. At 10 weeks of age, colon and ileum tissue was
   collected and used for RNA sequencing from mice that had been colonized
   with either responder (donor C) or non-responder (donor B) 6-week
   sample microbiota. There were 143 differentially expressed genes in the
   ileum and 16 in the colon (Fig. [194]6A, Supplementary Data [195]13).
   Gene ontology analysis found multiple pathways related to B-cell
   activation and immunoglobulin production, innate immune responses,
   inflammatory response, defense response to bacteria and response to
   lipopolysaccharide were enriched in the non-responder ileum tissue
   (Fig. [196]6B, Supplementary Data [197]14). The strong enrichment for
   B-cell pathways in non-responders included Ighg1, Ighg3 and Ighd
   antibody heavy-chain genes, but not IgA heavy-chain, as well as genes
   related to B-cell activation such as Cd40, suggesting that the type of
   antibodies secreted into the mucous layer and the lumen switches
   between responder and non-responder mice. KEGG enrichment identified
   several genes involved in tryptophan metabolism that were downregulated
   in responder mice including Cyp1a1, Kmo, and HAAO (Fig. [198]6C).
   Cyp1a1 is a negative regulator of tryptophan metabolite signaling via
   the aryl hydrocarbon receptor (Ahr), acting to degrade Ahr
   ligands^[199]25. KEGG enrichment also found NFκB signaling and cytokine
   receptor interactions as well as B cell receptor signaling were
   downregulated in responder mice (Fig. [200]6C and Supplementary
   Data [201]15). These data suggest that inflammatory mucosal immunity,
   in particular antibody production, is reduced by colonization with the
   microbiota-derived from SCFA-biotherapy responder donors.

Fig. 6. Ileal changes in B-cell, innate immunity, and tryptophan metabolism
gene signature in responder-microbiota colonized mice.

   [202]Fig. 6
   [203]Open in a new tab

   A RNA sequencing of ileum tissue from offspring of mice colonized with
   6-week samples from SCFA-biotherapy responder (donor C, n = 4) and
   non-responder (donor B, n = 4) mice, collected at 10 weeks of age.
   Volcano plot of differentially expressed genes in the ileum.
   Significance determined using Bioconducter package DESeq2 (two-sided
   analysis). Selected pathways enriched between responders and
   non-responders in the ileum using B Gene Ontology and C KEGG. Red
   labels: genes increased in responders, blue labels: genes decreased in
   responders. D CD40 expression on conventional dendritic cells (cDCs),
   CD11b + XCR1- dendritic cells (cDC2), total B cells, and IgM- B cells
   in the spleens of the responder (n = 7) and non-responder (n = 9) mice
   at 10 weeks of age. The gating strategy is shown in Supplementary
   Fig. [204]12. Data pooled from two experiments. The line represents
   median value. Significance was determined using an unpaired, two-tailed
   t-test. See also Supplementary Data [205]13–[206]15.

   To further investigate the activation status of immune cells in the
   mice, we analyzed antigen-presenting cell phenotypes in a group of
   10-week offspring. Consistent with the gene expression data, we found
   that CD40 expression on IgM^- B cells was higher in the mice colonized
   with 6-week sample microbiota from non-responders. In parallel to this,
   total conventional dendritic cells and type-2 dendritic cells had
   higher CD40 expression in the non-responder colonized mice. This
   suggests that both class-switched IgM^- B-cells and dendritic cells are
   more activated in the non-responder colonized mice.

Responder colonized mice have increased aryl hydrocarbon receptor ligand
production

   As the metabolism of tryptophan into Ahr ligands is a key mediator of
   intestinal homeostasis by the gut microbiota, we investigated Ahr
   ligand activity within the responder colonized mice. We stimulated
   Ahr-luciferase reporter cells with fecal water isolated from the fecal
   pellets of the colonized mice (Fig. [207]7A). We found that the fecal
   water of the responder mice had higher Ahr stimulatory activity than
   the non-responder mice, both in the baseline (p < 0.0001) and 6-week
   sample (p < 0.0001) colonized mice. Ahr stimulation activity increased
   in the 6-week sample colonized mice of both groups. We then used fecal
   waters from the human study participants and confirmed that Ahr ligand
   activity significantly increased at 6-weeks after the intervention
   (p = 0.0029, Fig. [208]7B). Of note, the two responder donors had
   higher Ahr activity than the non-responder donors at all three
   timepoints. Using immunofluorescence staining, we found that Ahr and
   IgA protein were both higher in the ileum tissue of the responder mice
   compared with the non-responder mice (Fig. [209]7C). This suggests that
   increased Ahr ligand availability after the SCFA-biotherapy is
   associated with diabetes protection.

Fig. 7. Enhanced aryl-hydrocarbon receptor activity in responder-microbiota
colonized mice.

   [210]Fig. 7
   [211]Open in a new tab

   A Fecal water extract from human-microbiota colonized mouse offspring
   at 7–10 weeks of age was used to stimulate Ahr luciferase reporter
   cells. Significance was determined by one-way ANOVA with Sidak’s
   correction (two-sided). Data was pooled from two experiments. B Fecal
   water extract from the human study participants (n = 18) was used to
   stimulate Ahr luciferase reporter cells. Human donors used in the mouse
   study were indicated by colors (non-responder donors – blue, responder
   donors – red). FICZ synthetic Ahr ligand was used as a positive control
   and data is expressed as fold change relative to unstimulated cells.
   Significance determined by repeated-measures one-way ANOVA with
   Dunnett’s correction (two-sided). C Representative images of ileum
   tissue from human-microbiota colonized mouse offspring at 10 weeks of
   age stained with anti-IgA (red), anti-Ahr (cyan), and DAPI (blue).
   Image representative of staining from n = 5 responder and n = 6
   non-responder mice from one experiment.

   To identify specific metabolites that may underlie the differences in
   Ahr activity, we re-interrogated our metabolomic datasets using an “in
   silico” targeted analysis of key tryptophan pathway metabolites
   (Supplementary Fig. [212]10). This analysis identified xanthurenic
   acid, a tryptophan metabolite with known Ahr stimulating activity
   downstream of kynurenine^[213]26, was significantly higher in the mice
   colonized with the 6-week responder donor microbiota (p = 0.0013,
   Supplementary Fig. [214]10A). In the human stool metabolome,
   2,8,dihydroxyquinoline and xanthurenic acid increased at 6 weeks and
   decreased at follow-up. Xanthurenic acid tended to be higher in the
   responder individuals at 6 weeks (Supplementary Fig. [215]10B). In the
   serum, tryptophan metabolites including tryptophan, kynurenine, and
   xanthurenic acid were all elevated at follow-up (Supplementary
   Fig. [216]10C). These data suggest that xanthurenic acid may be a key
   Ahr activating metabolite in both the mouse and human study.

Multi-omic factor analysis correlates microbiome markers, metabolites, stool
proteins, and aryl hydrocarbon receptor ligand production with glycemic
control measures

   We next used multi-omic factor analysis (MOFA) to understand the
   relationships between the different datasets in our human study (stool
   proteome and metabolome, plasma metabolome and stool metagenome,
   microbial taxa and pathways), clinical measures, SCFA and Ahr. The MOFA
   approach identifies latent factors comprising correlated variables
   selected from the different datasets in the model, which explain the
   underlying variation in the data (Supplementary Data [217]16)^[218]27.
   We then correlated these latent factors with clinical measures of
   glycemic control, SCFAs and Ahr activity. The MOFA model identified 6
   latent factors, with all five datasets contributing (Supplementary
   Fig. [219]11A). Latent factor 3 positively correlated with all three
   stool SCFAs and estimated C-peptide (Fig. [220]8A). Latent factor 3
   positive variables were mainly plasma metabolites, including the
   tryptophan metabolites indolelactic acid, kynurenine, tryptophan and
   indolepyruvate (Fig. [221]8B). Variables that negatively contributed to
   latent factor 3 included many stool metabolites such as pseudouridine,
   xanthine, oxoglutaric acid, and ornithine, which decreased after the
   biotherapy. Latent factor 4 positively correlated with stool and plasma
   butyrate and Ahr activity, while negatively correlated with HbA1c (%)
   and basal insulin (%). Latent factor 4 was driven by Bifidobacterium
   adolescentis and microbial pathways that increased at 6 weeks such as
   TCA cycle, glycolysis pathways and homolactic fermentation
   (Supplementary Fig. [222]11B)^[223]19. Latent factor 1 negatively
   correlated with stool Ahr activity and plasma acetate and was driven in
   the negative direction by taxa (Parabacteroides distasonis and
   Bacteriodes uniformis) and metagenomic pathways related to pyrimidine
   and purine metabolism (Supplementary Fig. [224]11C), variables which
   increased at 6-weeks. Latent factor 2 was negatively contributed to by
   similar taxa and also negatively correlated with Ahr activity, HbA1c
   and basal insulin % (Supplementary Fig. [225]11D). Together, this
   integrative analysis supports that SCFA, plasma tryptophan metabolites,
   and stool Ahr activity are all associated with clinical features of
   glycemic control in T1D.

Fig. 8. Multi-omic factor analyses (MOFA) identify latent factors associated
with glycaemic control, SCFA, and Ahr activity.

   [226]Fig. 8
   [227]Open in a new tab

   MOFA was used to integrate bacterial taxa (n = 22), bacterial pathways
   (n = 91), human stool proteins (n = 101), stool metabolite (n = 101)
   and plasma metabolite (n = 92) datasets. The identified latent factors
   were correlated with clinical data, SCFA and Ahr activity. A Heatmap
   indicating two-sided Pearson’s correlation between clinical variables
   and model factors. ***P < 0.001,**P < 0.01, *P < 0.05. B Variables
   contributing to latent factor 3. Only variables with absolute weight
   >0.4 are included in the plots. Black text: human proteins, blue text:
   plasma metabolites, red text: stool metabolites.

Discussion

   This is the first study to demonstrate remodeling of mucosal immunity
   and the gut barrier response after delivery of a potent SCFA-yielding
   intervention in adults with T1D. We demonstrated an increase in
   gut-derived antioxidant proteins, neutrophil repair and innate immune
   responses, epithelial and tight junction proteins, and amino acid
   metabolism along with a decrease in cell death pathways after delivery
   of a SCFA biotherapy. This was accompanied by a shift in fecal and
   circulating metabolites involved in the glutamate, arginine, butanoate,
   purine, pyrimidine, and tryptophan pathways. We then demonstrated that
   transplantation of the post-biotherapy fecal microbiota from butyrate
   responder participants delayed diabetes onset in germ-free NOD mice.
   Diabetes protection was associated with altered arginine, glutamate,
   purine, pyrimidine and tryptophan metabolism, similar to the pathways
   enriched in the human study. Within the mouse gut tissues of the
   diabetes-protected mice, IgG antibody, B-cell responses, and
   inflammatory innate immunity-related genes were downregulated while
   aryl-hydrocarbon receptor ligand activity and IgA were increased.
   Aryl-hydrocarbon receptor ligand activity and microbial markers
   correlated with measures of glycemic control in the patients. These
   data provide mechanistic insights into how a unique polysaccharide
   enriched with SCFAs acetate and butyrate, may interact with the host
   via intermediary metabolites to enhance mucosal immunity, impact the
   gut barrier, and protect from autoimmune diabetes.

   Using fecal proteomics allowed us to study the local host response to
   SCFA-biotherapy. Within the human proteins altered after the
   intervention, there was a strong signature related to neutrophils and
   myeloid cells as well as epithelial junction proteins and reduced
   cell-death pathways. Anterior gradient protein homolog-2 (AGR2), which
   increased after the intervention, protects goblet cells from
   endoplasmic reticulum stress and is required for the production and
   processing of gel-forming mucins^[228]28. Together, this suggests that
   the impact of neutrophils in the gut epithelium following the
   biotherapy is not cytotoxic or damaging to the gut epithelium.
   Neutrophil proteins that increased included MMP9 and CD11b, which are
   characteristic of tertiary neutrophil granules. Tertiary granules are
   much less cytotoxic than primary/azurophilic neutrophil granules and,
   in some circumstances can be regulatory and involved in resolution of
   inflammation^[229]29. Neutrophil-produced inflammatory proteins such as
   calprotectin, MPO and neutrophil elastase did not change after
   SCFA-biotherapy. MMP9 is a protease that facilitates transmigration of
   neutrophils into tissue^[230]30. Transmigrating neutrophils can enhance
   mucosal epithelium healing and resolution of inflammation via localized
   oxygen depletion^[231]31. Our pathway analysis showed an enrichment of
   antioxidant proteins related to response to hypoxia and cell redox
   homeostasis. Butyrate is known to stabilize hypoxia-inducible factors,
   aiding in maintaining a low-oxygen environment found in a healthy
   colonic epithelium^[232]32. High MMP9 expression is also characteristic
   of proangiogenic neutrophils, important for revascularization of
   hypoxic sites^[233]33. Restriction of oxygen availability in the gut is
   one mechanism the host uses to favor growth of beneficial anaerobic
   bacteria and restrict pathobionts^[234]34. Hence, regulation of oxygen
   levels in the intestinal epithelium may be an event downstream of
   SCFA-biotherapy, mediated by neutrophils and butyrate.

   N2 neutrophils are a subpopulation involved in tissue repair responses
   and are characterized by expression of MMP9, NOS2, ARG-1, TGF-β, and
   IL-10^[235]35. Polarization between pro - and anti-inflammatory immune
   subsets is heavily driven by arginine availability, with low arginine
   considered anti-inflammatory. NOS2 and ARG-1 are both downstream of
   ASS1 and part of the urea cycle and arginine metabolism pathways
   enriched in the human stool proteome, and metagenomics datasets. ASS1,
   the rate-limiting step in arginine biosynthesis, increased following
   SCFA-biotherapy and persisted after the washout period. Both NOS2
   (which was strongly increased at 6 weeks), and ARG-1 degrade arginine,
   reducing its bioavailability. N2 neutrophils, similar to M2
   macrophages, deplete arginine by converting it into NO, ornithine,
   proline, and polyamines, pathways involved in wound repair and
   regulating immunity^[236]36. Arginine starvation suppresses T cell
   proliferation via multiple mechanisms, including CD3ζ
   downregulation^[237]36,[238]37. Arginine was decreased in the
   diabetes-protected mice, and creatine, a product of arginine
   degradation that promotes M2 macrophage polarization^[239]38, was
   elevated in the human fecal metabolome. Production of NO via NOS2 is
   another immunosuppressive mediator that reduces T-cell activation via
   nitration of T-cell receptors and MHC molecules^[240]39,[241]40.
   Although excessive NO from the host, when converted to nitrate, can
   drive the expansion of facultative anaerobic bacteria contributing to
   dysbiosis^[242]41, we did not observe increases in dysbiotic
   bacteria^[243]19.

   Our data demonstrated that arginine metabolism in both the microbiota
   and host was activated following SCFA-biotherapy. In a large cohort
   study of children at risk for T1D (the TEDDY study), control subjects
   that did not progress to T1D or islet autoimmunity had a higher
   abundance of arginine and ornithine degradation, arginine biosynthesis
   and fermentation pathways within their microbiota^[244]5. In our mouse
   model, we observed lower fecal ornithine and arginine in the mice with
   delayed diabetes progression. In NOD mice, an inhibitor of conversion
   of arginine to ornithine by arginase-1 has been shown to protect
   against diabetes progression^[245]42. In vivo, ornithine is a major
   precursor for polyamines such as putrescine, spermidine, and spermine.
   Polyamines are increased in multiple autoimmune diseases and are
   important in T-cell immunity^[246]43. Within beta cells, polyamines are
   thought to contribute to endoplasmic reticulum and oxidative stress and
   play a role in islet inflammation^[247]44. Spermidine supplementation
   led to a higher incidence of diabetes in NOD mice^[248]45.
   Alpha-difluoromethylornithine (DFMO), an inhibitor of the enzyme that
   converts ornithine to polyamines, delayed the onset of experimental
   autoimmune encephalomyelitis^[249]46. In a recent phase I clinical
   trial in recent-onset T1D, DFMO showed preliminary evidence of
   preserving C-peptide at 6 months following treatment^[250]47 and is
   currently being trialed in a phase-2 study ([251]NCT05594563).

   The tryptophan metabolism pathway in the plasma was significantly
   enriched at follow-up. The gut microbiota transforms dietary tryptophan
   into indole derivatives like indolelactic acid and indolepyruvate.
   These compounds serve as ligands for the Ahr, a pivotal pathway in
   preserving intestinal equilibrium and immune tolerance^[252]48.
   Tryptophan is also metabolized by the host enzyme IDO to produce
   kynurenine, which contributes to the tryptophan pathway enrichment in
   plasma at follow-up. IDO activity was reported to be reduced in
   children with T1D^[253]49. Microbial producers of indolelactic acid
   include Parabacteroides distasonis, Bacteroides ovatus, and
   Bifidobacterium adolescentis^[254]50, all of which increased in our
   study participants after the intervention^[255]19. In our mouse model,
   tryptophan metabolites quinic acid and xanthurenic acid were higher in
   the feces of responder mice. Quinic acid is metabolized in the
   gastrointestinal tract leading to increased systemic tryptophan
   availability^[256]51. Quinic acid in serum was also reported to be
   decreased in children that rapidly progressed to T1D compared with slow
   progressors^[257]14. Xanthurenic acid is an Ahr ligand downstream of
   kynurenine, and usually considered an endogenous tryptophan
   metabolite^[258]52. However, certain gut bacteria can also metabolize
   kynurenine pathway intermediates into xanthurenic acid^[259]53.
   Xanthurenic acid is lower in inflammatory bowel disease patients, and
   xanthurenic acid reduced colitis symptoms and inflammation in animal
   models and promoted epithelial repair^[260]26. However, the role of
   these metabolites in type 1 diabetes remains to be explored.

   Gene expression analysis of the ileum tissue identified upregulation of
   Cyp1a1 in the diabetes-susceptible/non-responder group. While Cyp1a1 is
   upregulated by Ahr signaling, it acts as a negative regulator of Ahr by
   depleting Ahr ligands^[261]25. Our Ahr reporter assay demonstrated that
   the post-SCFA-biotherapy microbiota had higher Ahr stimulatory activity
   and that Ahr itself was upregulated in the ileum tissue. Administration
   of synthetic Ahr ligands in NOD mice have been shown to reduce
   insulitis and protect from diabetes via effects on both Foxp3^+ Tregs
   and RORγt^+ T cell populations^[262]54,[263]55. Similarly,
   administration of a diet yielding butyrate or acetate or a combination
   protected NOD mice from diabetes and islet autoimmunity^[264]4,[265]56,
   as well as protecting mice from gut infection^[266]57. With the
   combination of both acetate and butyrate metabolites attached to a
   polysaccharide biomolecule has a more potent and synergistic
   immunomodulatory effect for diabetes protection. Thus, our data
   indicate that SCFA-biotherapy may induce multiple beneficial metabolic
   pathways.

   Our analysis of the intestinal tissue indicated a potent upregulation
   of B-cell-related genes in the non-responder mice, including IghG1 and
   IghG3 antibody heavy chain genes, whereas IgA protein was increased in
   the responder ileum tissue. Microbiota-specific IgG production and FcγR
   activation by IgG is considered inflammatory within the intestinal
   mucosa and associated with ulcerative colitis, while mucosal IgA is
   associated with homeostasis^[267]58. Both acetate and butyrate have
   been shown to promote intestinal IgA production, in particular butyrate
   promotes lower-affinity, T-cell-independent IgA
   production^[268]57,[269]59,[270]60. The T-cell-independent IgA pathway
   tends to recognize conserved microbial antigens such as LPS, whereas
   T-dependent IgA leads to class-switching and high-affinity antibody
   responses. The enrichment of pathways such as “response to
   lipopolysaccharide”, “defense response to bacterium” and “inflammatory
   response” in the ileum tissue of the non-responder colonized mice
   suggests there was a qualitatively different environment that may have
   been promoting more potent immune activation. People with T1D had a
   higher proportion of IgA-coated bacteria, suggesting increased
   high-affinity IgA produced by class-switching, and this negatively
   correlated with SCFA levels^[271]7. In our proteomics analysis of the
   human stool samples, multiple antibodies changed in abundance across
   time, with some up and some downregulated. However, untargeted
   proteomics is not able to resolve the class of these antibodies. Thus,
   while butyrate production in our mouse model likely reduced mucosal IgG
   and increased IgA production, this mechanism still needs to be
   confirmed in human studies.

   Our study had several limitations related to the single-arm, open-label
   design and small sample size of our pilot human study. Our fecal
   proteomic analysis could not identify the source of the proteins
   detected, which may be derived from the any part of the intestine, the
   pancreas or another organ. Only proteins with a relatively high
   abundance were detected by our untargeted proteomics approach, and many
   biologically relevant low-abundance proteins will be missed. We also
   did not directly assess gut barrier function via intestinal
   permeability assessment in the study participants Likewise, our
   metabolomics analysis could be followed up with a targeted analysis for
   pathways of interest. While human fecal microbiota transfer is
   considered one of the best approaches for establishing disease
   causality^[272]61, this model has limitations. Only a proportion of the
   human microbiota colonized the mice, and those that did colonize
   changed in abundance compared to the donor samples, likely due to the
   different mouse diet, housing and physiology or preservation method of
   the stool samples. Some important taxa colonized the mothers but did
   not transmit well to the offspring (e.g., Faecalibacterium). However,
   we observed a functional replication in the mouse of the human donor
   butyrate response and metabolite profiles.

   We demonstrated that after undergoing SCFA-yielding biotherapy, adults
   with T1D show changes in the gut epithelium, mucosal immunity, and
   production of a range of metabolites. A functional shift in the gut
   microbiota drives enhanced amino acid and nucleotide metabolism both
   locally and systemically, particularly arginine, glutamate, purine,
   pyrimidine and tryptophan-related metabolites. Purine and pyrimidine
   metabolites correlated with markers of glycemic control. We were then
   able to link these changes in the human system with reduced diabetes
   progression using our mouse microbiome transfer model. Our analysis
   shows changes in the mucosal immune response after administering SCFAs
   were linked to the metabolism of purine, pyrimidine, arginine,
   glutamate, and tryptophan, as well as the anti-inflammatory effects of
   Ahr ligands. These changes were related to measures of glycemic
   control, suggesting a potential for clinical benefit. Here, we have
   uncovered that a specialized acetate-butyrate-yielding biotherapy
   positively impacts the production of beneficial downstream metabolites,
   indicating the wide-ranging potential of SCFAs in the slowing, halting,
   or reversing the progression of T1D.

Methods

Study participants

   Participants in this study were adults with T1D who had previously been
   described in a single-arm trial using an oral intervention consisting
   of a polysaccharide biomolecule modified with acetate and butyrate,
   also known as HAMSAB^[273]19, (Fig. [274]1A). The trial protocol was
   approved by the Sydney Local Health District Human Research Ethics
   Committee, the Human Research Ethics Committees of Monash University
   and University of Queensland (lead approval HREC/18/RPAH/241) and
   complies with all relevant ethical regulations. All patients provided
   written informed consent for the use of their samples and data in this
   study. The trial was registered with Australia New Zealand Clinical
   Trials Registry (ACTRN12618001391268). The trial participants were aged
   18–45 years, had a clinical diagnosis of T1D <6 months and HbA1c ≤ 8.5%
   (Supplementary Data [275]1). Estimated C-peptide was calculated as
   described using age, gender, HbA1c, insulin dose and body mass
   index^[276]62. All available stool and plasma samples were used in this
   study from the n = 21 participants collected at baseline (n = 20 stool
   and n = 18 plasma samples), 6 weeks (n = 18 stool and plasma samples),
   at the end of the supplementation period and 12 weeks (n = 18 stool and
   plasma samples), after a 6-week washout^[277]19. Additional approval
   for re-use of the participant stool samples for mouse colonization
   experiments was obtained from the University of Queensland Human
   Research Ethics Committee (low and negligible risk approval
   2023/[278]HE000166).

Mice

   NOD/ShiLtJArc (NOD) mice were originally obtained from the Animal
   Resource Facility (Perth, WA, Australia) and rederived into the
   Translational Research Facility Gnotobiotic Facility, within the
   Biological Resources Facility, University of Queensland. The germ-free
   NOD breeding colony was maintained on an autoclaved Teklad 2019S diet
   (Catalogue #2019S, Inotiv, Lafayette IN, USA) and all experimental mice
   were maintained on NIH07 autoclaved diet Catalog #7922, Inotiv) fed ad
   libitum. Sterile water was provided ad libitum. Germ-free mice were
   housed within sterile isolators with a 12-h light/dark cycle, 22 °C.
   67–72 median relative humidity. Following colonization with a complex
   human microbiota, experimental mice were housed in sterilized
   Tecniplast (Varese, Italy) individual isolator cages. Only female
   offspring of colonized mice were used for experiments. All experimental
   procedures were performed in sterile conditions within a biosafety
   cabinet. All experiments were approved by the University of Queensland
   Animal Ethics Committee (2021/[279]AE000553).

Human microbiota transfer into germ-free mice

   Baseline and 6-week stool samples from four trial participants were
   selected based on donor fecal butyrate response. Stool samples were
   processed within an anerobic chamber. Anaerobic glycerol solution
   (phosphate-buffered saline containing 12% glycerol and 0.1% Cysteine)
   was added to human stool and passed through a 70 μm cell strainer. The
   live bacterial cell count was determined using a BacLight Bacterial
   Viability Kit (Thermo Fisher Scientific, Waltham, MA). Germ-free NOD/Lt
   females (6 weeks of age) were gavaged on day 0 with 1.2 × 10^7 live
   bacteria (from a single donor sample) or sterile PBS and with
   2.4 × 10^7 live bacteria 7 days later. The colonized females were mated
   with germ-free NOD males 7-days after the second gavage. Female
   offspring were monitored weekly for diabetes onset starting at 10 weeks
   of age via blood glucose testing from tail-vein blood collected by
   sterile pipette and tested with a Freestyle Optium Blood Glucose
   Monitoring device (Abbott). Mice were deemed diabetic after two
   consecutive readings of >16 mmol/L. The stool was collected from
   mothers at week 2 and from offspring at 5 and 11 weeks for 16s rRNA
   gene amplicon sequencing analyses and at weeks 7 and 9 for metabolite
   analyses.

Stool protein extraction, digestion and purification for proteomics

   Human stool samples were resuspended in 3:1 (w/v) of ice-cold potassium
   phosphate buffer (50 mM containing 5 mM EDTA). Heavy particles were
   allowed to settle by incubating on ice for 30 min and the supernatant
   retained. Bacteria and debris were pelleted by centrifuging at 8000 g
   for 10 min at 4 °C. The soluble fraction, which is enriched for human
   proteins^[280]20, was then removed and re-spun at 8000 g for 10 min at
   4 °C and the supernatant was used for both proteomics and metabolomics
   analysis (see below). Isolated protein (20 μg as estimated by the
   direct detect spectrometer), was lysed in a solution containing 1%
   sodium deoxycholate, 10 mM tris[2-carboxyethyl]phosphine and 40 mM
   2-chloroacetamide in 100 mM Tris pH 8 and boiled at 95 °C for 5 min.
   Proteins were digested using porcine trypsin (Promega, Madison, WI) at
   a protein:trypsin ratio of 50:1 (w/w) for 16 h at 37 °C. Protein
   digestion was halted and detergent precipitated using trifluoroacetic
   acid (TFA) at a final concentration of 5%. Samples were clarified by
   centrifugation at 13,000 × g. Tryptic peptides were desalted using
   Glygen C18 pipette tips on an Agilent Bravo liquid handler. The eluent
   was evaporated with centrifugation under vacuum and tryptic peptides
   were resuspended into a final concentration of 0.5 μg/μL with 0.5% TFA
   for mass spectrometry.

Proteomic mass spectrometry and protein identification

   Samples (n = 56) were analyzed in duplicate on a Q Exactive High-Field
   mass spectrometer coupled with an UltiMate 3000 UHPLC (Thermo Fisher
   Scientific). Tryptic peptides were loaded on a Thermo Fisher Scientific
   trap column 160454 (0.3 mm × 5 mm) and separated with a Thermo Fisher
   Scientific EASY-Spray HPLC column ES803 (75 μm × 500 mm), over a
   100-min gradient from 3 to 25% acetonitrile over 60 mins followed by
   25–40% acetonitrile over 12 min in buffer containing 0.1% formic acid
   at a flow rate of 250 nL/min. The columns were washed with 95%
   acetonitrile for 15 min and re-equilibrated with 3% acetonitrile for
   7 mins. Survey scans were acquired from 375m/z–1500m/z with a
   resolution of 60,000, automatic gain control of 3 × 10^6, and maximum
   injection time of 50 ms. Precursor ions consisting of the top 20 most
   intense ions were selected for tandem mass spectrometry and fragmented
   with a normalized collision energy of 28%. Dynamic exclusion was set to
   20 s, and the mass exclusion width of 10 parts per million. Second
   stage of mass spectrometry was acquired with between 200m/z–2000m/z at
   a resolution of 15,000, automatic gain control of 5 × 10^5, and maximum
   injection time of 20 ms.

   Human peptides and proteins were identified and quantified using
   Proteome Discoverer (v2.5.0.400, Thermo Fisher Scientific), which
   applies label-free quantification methods. Search parameters included
   carbamidomethylating of cysteine and deamidation of asparagine as fixed
   and variable modifications respectively. A maximum of two missed
   cleavages were allowed and the false discovery rate was set at 0.01.
   Proteins were searched against the UniProt human database. Common
   contaminants identified by MaxQuant (v2.3.2.0, Max-Planck-Institute of
   Biochemistry) were removed.

Proteomics data processing

   Protein intensities from duplicate samples were averaged. Proteins that
   were not present in a minimum of 40% of a single timepoint were
   excluded. A multiple mixed imputation strategy was implemented using R
   package, imp4p^[281]63. Here, the likelihood of a protein to be
   “missing completely at random” (MCAR) or “Missing not at random” (MNAR)
   is predicted using Bayesian probabilities. Proteins identified as MNAR
   were imputed using ‘Imputation under a Gaussian Complete Data
   Assumption’, and those determined as MCAR were imputed using random
   forest. Intensities were subsequently log2 transformed and normalized
   using variance stabilization (R package “vsn”) before linear mixed
   model analysis using R package ‘lme4’.

Fecal and plasma metabolite sample preparation

   Metabolites were analyzed from plasma (25 μL, N = 54), mouse fecal
   pellets (0.2 g, N = 39) or the enriched human soluble stool fraction
   (N = 56) prepared for proteomics (see above). Plasma samples were
   prepared at 4 °C by the addition of ice-cold 200 µL of 1:1
   acetonitrile:methanol was added and mixed for 20 min. Subsequently,
   samples were centrifuged at 22,000 × g for 10 min and the supernatant
   remove for further analyses. For mouse fecal pellets, 1 mL of 50%
   acetonitrile was added and homogenized with metal beads for 10 min at
   30 Hz (Retsch Mixer Mill MM400) at 4 ^oC. Homogenate was then
   centrifuged at 16,000 × g for 10 min at 4 ^oC and the supernatant
   removed and diluted with 50% acetonitrile until the concentration was
   2 mg dry weight/mL. Mouse fecal homogenate was used for untargeted
   metabolomics and SCFA quantification (see below). The enriched human
   soluble stool fraction only required protein and salt removal. Proteins
   and salts were removed from human fecal supernatants, mouse fecal
   homogenates or plasma by adding cold extraction solvent containing
   methanol spiked with the fully isotopically labeled internal standards
   (1:1000) 13 C,15N-amino acid mixture (Cell Free Amino Acid Mixture -
   13C,15N, Merck, Rahway, NJ, USA), 1 μM of buffers
   N-cyclohexyl-3-aminopropanesulfonic acid, (3-((3-cholamidopropyl)
   dimethylammonio)-1-propanesulfonate,
   piperazine-N,N′-bis(2-ethanesulfonic acid) and 2 μM of the antioxidant
   butylated hydroxytoluene, and incubating at 4 °C for 1 h. Following
   centrifugation at 14,000 × g for 10 min, supernatants were transferred
   to an LC/MS vial for analysis and 10 μL injected into LCMS (1 injection
   per sample).

Metabolomics LC-MS analysis

   LCMS analysis was performed by the Monash Proteomics and Metabolomics
   Platform and acquired on Q-Exactive Orbitrap mass spectrometer (Thermo
   Fisher Scientific) coupled with a Dionex Ultimate® 3000 RS HPLC system
   (Thermo Fisher Scientific). Chromatographic separation was performed on
   a ZIC-pHILIC column (5 µM, polymeric, 150 × 4.6 mm, SeQuant®, Merck)
   with buffer A as 20 mM ammonium carbonate and buffer B acetonitrile,
   with HPLC and MS settings as previously described^[282]64. Flow rate
   was set to 0.3 mL/min, gradient was as follows: 0 min 80% B, 15 min 50%
   B, 18 min 5% B, 21 min 5% B, 24 min 80% B and 32 min 80% B. Column
   compartment temperature was 25 °C and autosampler temperature was 4 °C.
   Heated electrospray ionization (HESI) source parameters were: sheath
   gas 50, aux gas 20, sweep gas 2 arbitrary units, capillary temperature
   300 °C, S-lens RF level 50, aux gas temperature 120 °C, spray voltage
   4.0 kV in positive mode, 3.5 kV in negative mode. MS data was collected
   in full scan mode with positive and negative polarity switching over 85
   – 1275 m/z range, at 35k resolution, with 1e6 AGC and 50 ms maximum
   injection time. A pooled sample was used to acquire MS2 data for ID
   purposes in DDA mode in positive and negative ion detection modes as
   separate injections. Samples were analyzed in randomized order and
   pooled quality control samples were analyzed throughout each batch to
   confirm method reproducibility. Mixtures of pure authentic reference
   standards containing over 320 metabolites^[283]64 were acquired as
   separate injections before each study and used to confirm retention
   times to assist with metabolite identification.

Metabolomics data processing

   The acquired LCMS data was processed in untargeted fashion using open
   source software IDOEM^[284]65. The standard IDEOM workflow with default
   parameters was used to detect and group peaks, eliminate unwanted noise
   and artefacts, and identify and quantify metabolites, as previously
   described^[285]64. Metabolites were identified by searching against the
   IDEOM database based on accurate mass and retention time (from
   authentic standards analyzed in the same batch) for Metabolomics
   Standards Initiative level 1 metabolites. Where authentic standards
   were not available, putative metabolite identities are annotated
   (Metabolomics Standards Initiative level 2) based on predicted
   retention time^[286]66. Annotated metabolites were inspected for peak
   quality and the following additional data filters were applied: When a
   metabolite was identified more than once, mass and retention times were
   inspected to select the most likely peak corresponding to that
   metabolite (based on chromatographic behavior of the authentic
   standards). Metabolites with an average peak intensity of less than
   500,000 were removed to reduce the likelihood of LC-MS artefacts being
   misidentified as metabolites. Metabolites that were not present in at
   least 40% of samples for a single timepoint were removed. Metabolites
   with less than two missing values were imputed with the average value
   from the timepoint. The remaining metabolites were log2 transformed and
   normalized using variance stabilization methods in the R package ‘vsn’
   before linear mixed model analysis using R package ‘lme4’, similar to
   proteomic samples. Correlations between stool metabolites that were
   significantly altered after the SCFA biotherapy (FDR < 0.01) and
   glycaemic variables estimated C-peptide and glycated hemoglobin (Hb1Ac)
   was performed using repeated measures correlation (R package rmcorr).
   Correlation p-values were adjusted for multiple correction using FDR.

   Peak areas of tryptophan pathway metabolites were obtained by searching
   metabolomics raw data using the Tracefinder application (Thermo Fisher
   Scientific). Tryptophan metabolites that were detected and integrated
   were: 2,8-dihydroxyquinoline, indole-3-propionic acid, indole acetic
   acid, indole lactic acid, kynurenine, tryptophan and xanthurenic acid.
   Identities of these features were confirmed with reference standards or
   MS2 spectrum and 3 ppm mass error cut off was used. Other metabolites
   searched for but not found: 1H-indole-3-carboxaldehyde,
   3-methyl-2-oxindole, 3-methylindole, indole,indole-3-acrylic acid,
   indole-3-carboxylic acid, kynurenic acid. Spectral identification
   information for tryptophan metabolites is provided in Supplementary
   Data [287]18.

SCFA quantification on mouse fecal samples

   A liquid chromatography-multiple reaction monitoring mass spectrometry
   (LC/MRM-MS) method measuring 76 transitions from 11
   nitrophenylhydrazine (NPH)-derivatised SCFAs (Supplementary
   Data [288]18) was developed based on^[289]67,[290]68. Method
   development is detailed in the Supplementary Methods. Mouse feces (2 mg
   dry weight/mL), calibration standards, blank and a pooled quality
   control sample were derivatised with 3-NPH, and then mixed with an
   isotope-labeled internal standard solution prepared using
   ^13C[6]-3-NPH·HCl (Cayman Chemical, Ann Arbor, Michigan) prior to LC-MS
   analysis in negative ionization mode on a 1290 Infinity II UHPLC
   coupled to a 6470 QQQ mass spectrometer (Agilent, Santa Clara, USA).
   Samples (n = 108) were analyzed in a randomized order bracketed by
   quality control and calibration standards. Control groups included
   n = 5 germ-free NOD mice and n = 10 SPF NOD mice. Skyline^[291]69 was
   used for peak picking, integration, and calibration curve construction,
   while downstream calculations were performed in Excel. SCFA transitions
   are shown in Supplementary Data [292]18.

Mouse stool DNA extraction and 16 s rRNA library preparation

   DNA from mouse feces was extracted using 3:1 ratio of 0.1 mm+1 mm
   Zirconia beads (Daintree Scientific) in lysis buffer (50 mM Tris-HCl,
   500 mM NaCl, 50 mM EDTA, 4% SDS). Samples were homogenized using a
   Precellys 24 (Bertin Technologies) and then supernatants then treated
   with RNAse and then Proteinase K. DNA was subsequently extracted using
   the automated Maxwell 16 LEV Blood DNA extraction kit (Promega,
   Madison, WI). Variable region V6-V8 were amplified using the
   Illumina-seq-926F (AAACTYAAAKGAATTGRCGG) and 1392R2 (ACGGGCGGTGWGTRC)
   primers with Nextera XT (Illumina) adapter sequences^[293]70. Amplicon
   sequences were purified using E-gels. Indexed amplicon libraries were
   then purified using AMPure XP beads (Beckman Coulter, Brea, CA, USA),
   pooled to an equimolar concentration of 10 nM. Sequencing was performed
   by the Australian Centre for Ecogenomics (University of Queensland,
   Brisbane, Australia) using an Illumina MiSeq using the MiSeq V3
   chemistry (2x 300 bp) according to the manufacturers protocol.

16s Amplicon data processing

   16s rRNA amplicon sequences were demultiplexed and preprocessing
   performed using ‘^[294]71. Primer sequences were removed from forward
   de-multiplexed reads using q2-cutadapt plugin. Amplicon sequences were
   denoised and trimmed to 250 bp using dada2 plugin. Taxonomy was
   assigned using a classifier trained on V6 – V8 16 s region of the 16S
   SILVA database (release 138, clustered at 99% identity). Generated
   amplicon sequence variant taxonomic tables were analyzed and compared
   at the genus level. Taxa present in two or less mice or had a minimum
   abundance <0.01% were removed from analyses. Bacterial counts were
   rarefied at 15,000 and normalized using center log-ratio with an offset
   of 1. Differences in microbiome taxonomic composition between responder
   groups from donor 6-week trial stool samples were determined using
   ANOVA.

RNA sequencing on mouse intestinal tissue

   Colon and ileum were collected in RNAlater Stabilization Reagent
   (Thermofischer Scientific). Total RNA was extracted with RNeasy Mini
   Kit (Qiagen). Transcriptomics sequencing analysis was performed by
   Azenta Life Sciences using Illumina paired end 2×150 base pair
   configuration using an Illumina NovaSeq platform to a depth of 79-95
   million reads per sample. Raw transcripts were inspected for quality
   control using FastQC (version 0.10.1) with the average quality score
   for each transcript above 30. Adapter sequences and sequences with <75
   bases were removed using Cutadapt (version 1.9.1). Transcripts were
   aligned to the reference genome Mus Musculus using Hisat (version
   2.01).

Flow cytometry

   Single-cell suspensions were prepared from spleen and lymph nodes by
   passing cells through a 70-micron cell strainer (BD Biosciences,
   Franklin Lakes, NJ). Red blood cells in spleen were lysed with ACK
   lysing buffer. Cells were stained for surface and/or intracellular
   markers in the dark at 4°C. T-cells were phenotyped using anti-mouse
   antibodies CD8-BUV805 (Clone 53-6-7, 1:400), CD4-BUV496 (GK1.5, 1:400),
   CD44-BUV395 (IM7, 1:400) from BD Biosciences, CD279 (PD-1)-BV711
   (29 F.1A12, 1:400), CD152(CTLA4)-BV605(UC10-4B9, 1:200) from
   (BioLegend, San Diego, CA, USA), FoxP3-PE-Cy5 (FJK-16s, eBioscience,
   1:400). I-A^g7- BDC2.5mim monomers specific for islet peptide were from
   the NIH tetramer core facility and tetramerized with streptavidin-PE
   (Prozyme, Seattle, WA, USA) as described^[295]72. Dead cells were
   excluded with FVS440 for live/dead stain from BD Biosciences. Antigen
   presenting cells were phenotyped using anti-mouse antibodies CD80-PE
   (16-10A1, 1:800), CD40-PE-Cy7 (3/23, 1:400), CD11c-BV605 (N418, 1:200),
   I-A^d-AF647 (39-10-8, 1:400), H-2Kd-FITC (SF1-1.1, 1:200), XCR1-BV650
   (zet, 1:200), TCR-APC Fire750 (H57-597, 1:400), CD19-PerCP-Cy5.5 (6D5,
   1:200), CD274 (PDL1)-BV785 (10 F,9G2, 1:200) from Biolegend.
   CD11b-BUV395 (M1/70, 1:200), CD86-BV510 (GL1, 1:200), IgM-BV421(UCH-B1,
   1:200) from BD Biosience and Aqua for dead cell exclusion (Thermofisher
   Scientific, Waltham, MA, USA). Flow cytometry was performed using a BD
   FACSymphony A5 using FACS DiVa 8.0.1 software and data analyzed using
   FlowJo V10.8.1 software (Tree Star, Ashland, OR, USA). Gating strategy
   is shown in Figs. [296]S6 and [297]S12.

Ahr reporter cell assay

   HT29 cells (ATCC HTB-38) were transduced with lentivirus carrying Ahr
   with firefly luciferase reporter (Ahr vector # 90337, with enveloping
   and packaging vectors pCMV-VSV-G #8454 and pCMV-dR8.2 dvpr # 8455 all
   from addgene, Waterton MA, USA). HT29 Ahr reporter cells (50,000) were
   plated on white 96 well plate in complete media (DMEM supplemented with
   10% FBS, 1% Penicillin Streptomycin, and 1% Glutamine) and incubated at
   37 °C overnight. Next day, cells were treated with 10% v/v of fecal
   water (as prepared for proteomics) for 4 h.
   6-Formylindolo[3,2-b]carbazole (FICZ, 100 nM, Sigma-Aldrich) was used
   as a positive control. Luciferase assay was performed using Pierce™
   Firefly Luciferase Glow Assay kit (Thermofisher). Ahr activity was
   normalized to cells only control and presented as fold change.

Insulitis and immunofluorescence staining

   Pancreata were fixed in 10% neutral buffered formalin (Sigma-Aldrich,
   St Louis, MO, USA) and stained using hematoxylin and eosin. Slides were
   scanned using an Olympus VS120. Insulitis were assessed as follows: 0 =
   no infiltration, 1 = peri-insulitis, 2 = mild insulitis (<50% of islet
   infiltration), 3 = severe insulitis (>50% of islet is affected by
   pathological change), 4 = complete infiltration. A minimum of 30 islets
   were scored per pancreas using blinded analysis. For
   immunofluorescence, pancreas sections were stained with anti-insulin
   (Clone #182410, 1:2000, from R&D Systems, Minneapolis, MN, USA),
   anti-glucagon (Clone ICACLS, 1:500, from eBiosiences). Secondary
   antibodies donkey anti-rat AF488 and donkey anti-mouse AF647
   (Invitrogen, used at 1:500) and DAPI to further identify residual
   islets. Slides were scanned on Olympus VS200 and exported to Image J
   for analysis.

   Ileum and colon tissue were rolled and fixed in 10% neutral buffered
   formalin before processing in paraffin blocks. Paraffin sections were
   rehydrated then antigens retrieved with citrate buffer. Sections were
   permeabilized with 0.5% Triton-100, followed by serum blocking. Primary
   antibodies were goat polyclonal anti-mouse IgA-UNLB (1:500) and Ahr
   monoclonal (Clone RPT1, 1:150) antibody, and secondary antibodies were
   Donkey anti Goat Alexa Fluor®568 and Donkey anti Mouse Alexa Fluor®647
   (both used at 1:500) with DAPI nuclei staining. Fluorescence images
   were acquired using Confocal Microscope Olympus FV3000.

Statistical analysis

   Changes in multivariate signatures of each ‘omic dataset across
   timepoints were investigated using supervised Partial-Least-Squares
   Discriminant-Analyses (PLS-DA) and unsupervised Principal Component
   Analyses (PCA) using the R software package, mixOmics
   (V6.23.2)^[298]73. Significance was determined using Permutation
   Analysis of Variance adjusted for repeated measures (PERMANOVA) using
   the R package vegan (V2.6-6)^[299]74. Pairwise differences were
   determined using pairwiseAdonis package (V0.4) and adjusted for
   multiple comparisons using false discovery rate. Significant
   differences in individual stool proteins, plasma and stool metabolites
   were inferred using linear mixed models in the R package lme4
   (V1.1-31). Timepoints was accounted as a random factor. Statistical
   significance was defined based on Likelihood Ratio Test (p < 0.05),
   with Benjamini-Hochberg correction for multiple testing. Tukey HSD test
   determined pairwise differences between timepoints^[300]75. Proteins
   with global p < 0.05 were explored for pathway enrichment (KEGG and
   GO:Biological Processes). Separate pathway enrichment analyses were
   performed on proteins that either significantly increased or decreased
   (p < 0.05) from baseline to six weeks or baseline to follow-up using
   the software ShinyGo V0.77^[301]76, with an FDR cut off of 0.05. All
   protein coding genes were used as background. Fecal metabolite pathway
   enrichment was compared using MetOrigin (V1.0)^[302]77. Differences
   were determined using paired t-test of individual metabolites with a
   cut-off value of p < 0.05. Metabolite Pathway Enrichment Analyses
   (MPEA) was determined using an ‘origin’ based approach where identified
   metabolites are searched against three databases: host-only (human KEGG
   pathways), microbial only (all KEGG bacterial pathways) or combined
   host/microbe. Identified metabolites are compared with all other
   pathway associated metabolites in each origin-based database.
   Foldchange analyses confirmed directional changes of significantly
   different metabolites between groups. Plasma metabolite enrichment
   analysis was conducted using web-based MetaboAnalyst v5.0. Significant
   metabolites identified under linear mixed models (p < 0.05) were
   selected for the enrichment analysis (mapping to the pathway-based KEGG
   metabolite set library).

   RNA-seq gene expression was normalized using fragments per kilo bases
   per million reads (FPKM). Differential gene analysis and pathway
   ontologies enriched in ileum and colons tissues of non-responder and
   responder mice were analyzed separately. Gene differential analyses on
   read count data was performed using Bioconductor package DESeq2
   (V1.6.3) based on a model with a negative binomial distribution.
   Functional gene ontology analyses were performed using GOSeq with a p
   value threshold of 0.05^[303]78. Differences between stool SCFA and Ahr
   activity in humanized mice responders and non-responders from the
   different trial donor samples were compared with ANOVA and diabetes
   survival curves were compared using log-rank (Mantel-Cox) survival
   analysis using using Graphpad Prism v10.4.1.

Multiomic factor analysis

   Multiomic Factor Analysis (MOFA) was performed using the MOFA2 (V3.20)
   R package^[304]27. Bacterial taxa, bacterial pathways, human stool
   proteins, plasma metabolites and stool metabolites were incorporated as
   separate “views”. Bacterial taxa and pathways were from our previously
   published metagenomics anlaysis^[305]19. Variables in each dataset with
   p.value < 0.05 were selected for analyses. Each “view” was normalized
   as stated in the respective univariate analyses. Prior to model
   training, within sample variation was removed. MOFA analysis was
   performed with 1 analysis group using default setting except “ard
   factors” were set to “true” to account for the sparsity of the views.

Reporting summary

   Further information on research design is available in the [306]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [307]Supplementary Information^ (6.2MB, pdf)
   [308]41467_2025_58319_MOESM2_ESM.pdf^ (55.5KB, pdf)

   Description of Additional Supplementary Files
   [309]Supplementary Data 1–8^ (102KB, xlsx)
   [310]Supplementary Data 9–18^ (204.2KB, xlsx)
   [311]Reporting Summary^ (121.4KB, pdf)
   [312]Transparent Peer Review file^ (217.7KB, pdf)

Acknowledgements