Abstract

   The human gut microbiota has the potential to synthesize proteins that
   may influence host metabolism. Here we report two polypeptides,
   RUMTOR-derived peptide (RORDEP) 1 and RORDEP2, circulating in human
   blood and synthesized by specific strains of gut commensal Ruminococcus
   torques that correlate inversely with adiposity in humans. Oral gavage
   with RORDEP-expressing strains improved glucose tolerance, increased
   bone density and reduced fat mass with an enhanced expression of genes
   and proteins involved in thermogenesis and lipolysis in lean mice on a
   high-fat diet and diet-induced obese mice. Recombinant RORDEP1 given to
   rats intraperitoneally decreased plasma gastric inhibitory polypeptide
   but increased glucagon-like peptide 1, peptide YY and insulin.
   Intestinal delivery of recombinant RORDEP1 to rats potentiated
   insulin-mediated inhibition of hepatic glucose production by
   downregulating genes and proteins controlling liver gluconeogenesis,
   glycogenolysis and lipogenesis but upregulating those involved in
   insulin signalling, glycogenesis and glycolysis. These preclinical
   findings warrant the exploration of RORDEPs for the prevention and
   treatment of human metabolic disorders.

   Subject terms: Endocrine system and metabolic diseases, Endocrine
   system and metabolic diseases
     __________________________________________________________________

   Two polypeptides synthesized by common bacterial strains in human gut
   microbiota lower body fat and blood glucose and increase bone density,
   improving the metabolism of rodents.

Main

   Human biology is contextual on trillions of various microorganisms that
   interact with their host^[78]1. Central to this interaction is the
   collection of microbial genes, the microbiome, which in a human
   individual may represent a repertoire of protein-coding genes that is
   more than one order of magnitude larger than the number of genes in the
   human nuclear genome^[79]2,[80]3.

   Despite the daunting task of annotating this multitude of microbial
   genes, large-scale computational mapping of thousands of human
   microbiome genes and their predicted proteins has been
   initiated^[81]2,[82]4–[83]7. Many of the annotated proteins have
   functions within their microbial communities including housekeeping,
   microorganism–microorganism communication, adaptation and defence
   against other microorganisms. In addition, other specific microbial
   compounds have been mapped including antimicrobial peptides,
   antibiotics, and proteins or lipids that in preclinical and for a few
   cases also in clinical settings modulate host biology^[84]8–[85]17.

   Among host functions that are influenced by gut microbiota, those
   involved in metabolism are of particular interest for developing novel
   approaches to prevent or treat widespread human metabolic disorders
   such as obesity, diabetes and osteoporosis^[86]18. Ruminococcus torques
   (RT) is a prevalent and abundant member of the commensal human gut
   microbiota^[87]19, yet its functional role in host metabolic regulation
   remains largely unexplored.

   Here we tested the hypothesis that gut bacteria coexisting with humans
   produce and release proteins that impact host metabolism^[88]20,[89]21.
   We identified multiple strains of RT, which are predicted to harbour a
   known protein, RUMTOR_00181, the structure of which includes two
   fibronectin type III (FN3) protein domains. The two FN3 protein domains
   of RUMTOR_00181 share 73% amino acid residue identity, and they have
   24% and 25% amino acid residue identity with irisin, a known human
   polypeptide hormone^[90]22. We have named the two bacterial FN3 protein
   domains RUMTOR-derived peptide (RORDEP) 1 and RORDEP2. In human
   epidemiological studies, we show that the abundance of RT strains
   expressing RUMTOR_00181 is inversely correlated with body mass index
   (BMI) and body adiposity. Following interventions in mice with an RT
   strain expressing both RORDEPs or an Escherichia coli strain engineered
   to express RORDEP1, we find an improvement of glucose tolerance, a
   higher bone density and a reduction of weight gain driven by a high-fat
   diet. Culture experiments suggest that the two RORDEPs are released
   from a RUMTOR_00181-expressing RT strain into culture supernatants and
   are present in human blood. Intraperitoneal delivery of recombinant
   RORDEP1 (r-RORDEP1) in rodents improves glucose tolerance and, in
   response to an oral glucose load, elicits an increase of plasma
   glucagon-like peptide 1 (GLP1), peptide YY (PYY) and insulin but a
   decline in gastric inhibitory polypeptide (GIP). Following intestinal
   delivery to rats, r-RORDEP1 lowers blood glucose by a potentiation of
   insulin-mediated inhibition of liver glucose production.

Results

Some RT strains are predicted to synthesize RORDEPs

   We undertook a computational search by aligning RefSeq^[91]23
   prokaryote representative genomes containing 285,952 sequences (as of
   17 October 2018) stored at the National Center for Biotechnology
   Information (NCBI) with amino acid sequences annotated to 118 human
   ligands and their precursor proteins (Fig. [92]1a and Supplementary
   Table [93]1). Alignment criteria are shown in the illustration of the
   bioinformatics pipeline in Fig. [94]1a.

Fig. 1. Identification of RUMTOR_00181 and RORDEPs.

   [95]Fig. 1
   [96]Open in a new tab

   a, Utilization of a bioinformatics-driven genomic alignment approach
   revealed the presence of a RUMTOR_00181 protein in RT strains of human
   gut microbiota. The bacterial protein contains two FN3 protein domains
   with modest identity to the human FNDC5. b, Predictive modelling of the
   bacterial RUMTOR_00181 protein structure as performed in the AlphaFold
   Protein Structure Database
   ([97]https://alphafold.ebi.ac.uk/entry/A5KIY5); this also holds for the
   structural prediction of RORDEPs. The AlphaFold-predicted 3D structure
   was visualized using Pymol ([98]https://pymol.org/2/). SP, signal
   peptide; TD, transmembrane domain. c, A schematic representation of the
   topology of RUMTOR_00181 protein. aa, amino acids. The putative trypsin
   cleavage sites were predicted using PeptideCutter
   ([99]https://web.expasy.org/peptide_cutter/). Panel a created with
   [100]BioRender.com.

   We find a significant similarity between a sequence annotated as human
   irisin and its precursor protein FN3-containing protein 5 (FNDC5), and
   a coding sequence annotated as RUMTOR_00181 (UniProt: [101]A5KIY5) in
   the genome of the RT Amercian Type Culture Collection (ATCC) 27756
   strain isolated from human microbiota. The bacterial genome has been
   deposited at NCBI as the reference genome for the RT species.

   The identified RUMTOR_00181 is predicted to have a signal peptide, two
   FN3 sandwich structure domains and one hydrophobic domain, the latter
   probably needed for membrane insertion, followed by a seven-amino acid
   C-terminus (Fig. [102]1b,c and Supplementary Fig. [103]1). Moreover,
   within the 244 available metagenomic or culturomic RT strain genomes,
   we find 161 strains isolated from Homo sapiens with a high homology
   (>99%) to RUMTOR_00181 from RT ATCC 27756 (Supplementary Table [104]2).

RORDEPs are released in cultures of RT expressing the precursor protein

   RUMTOR_00181 is predicted to be proteolytically cleaved by trypsin-like
   endopeptidases present within the genome of the RT ATCC 27756 strain
   (Supplementary Table [105]3) as well as in the intestinal lumen into
   two polypeptides. These we named RORDEP1 and RORDEP2, consisting of 87
   and 88 amino acid residues, respectively (Fig. [106]1c). Phylogenetic
   analysis comparing RORDEPs with 1,077 sequences of known FN3-containing
   human proteins shows that both RORDEP1 (Supplementary Fig. [107]2a) and
   RORDEP2 (Supplementary Fig. [108]2b) have homology to FNDC5, with
   identities of 13% (28 out of 214 amino acid residues for RORDEP1 and 27
   out of 214 amino acid residues for RORDEP2; Supplementary Fig. [109]3).
   The alignment of RORDEP1 and RORDEP2 shows a 73% (65 out of 89 amino
   acid residues) identity (Supplementary Fig. [110]3).

   To experimentally validate the computational predictions, we performed
   liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS).
   The analysis focused on identifying RORDEPs in culture supernatants of
   RT ATCC 27756. As shown in Extended Data Fig. [111]1, for the mass
   spectrometric analysis of each RORDEP, two unique proteotypic peptides
   were chosen as markers to unambiguously identify RORDEP1 and RORDEP2.
   Given the molecular weights of RORDEPs (approximately 10 kDa), we
   selected the 3–30-kDa fractions from culture supernatants of RT ATCC
   27756 synthesizing RUMTOR_00181 to develop the assays (Supplementary
   Fig. [112]4). The presence of RORDEPs in culture media was confirmed by
   the high mass accuracy of the b- and y-ion series in the MS^2 spectra,
   comparing the predicted and observed data for the unique proteotypic
   peptides (Extended Data Fig. [113]1 and Supplementary Table [114]4).
   The presence of RORDEP1 and RORDEP2 in the RT ATCC 27756 culture
   supernatants therefore aligns with the predicted proteolytical cleavage
   and release of the two polypeptides from the RT ATCC 27756 strain as
   fragments with a molecular size below 30 kDa.

Extended Data Fig. 1. Proteomics identification of selected proteotypic
peptides that are unique to RORDEPs in supernatants of bacterial cultures
using liquid chromatography-mass spectrometry (LC-MS).

   [115]Extended Data Fig. 1
   [116]Open in a new tab

   In each panel, the left section displays the selected proteotypic
   peptides (underscored and highlighted in red) that are unique to
   sequences of RORDEPs. In the central section, the left graph displays
   the fragmentation patterns for the unique peptides in the
   identification of RORDEPs, with b- and y- ions indicating that the
   fragmentation occurs in a way that the charge is retained on the
   N-terminus and C-terminus, respectively, of the respective peptide. The
   b-ions contain the amino-terminal end of the peptides and are numbered
   starting from the N-terminus (the beginning of the peptide chain). The
   y-ions contain the carboxyl-terminal end of the peptide and are
   numbered starting from the C-terminus (the end of the peptide chain).
   The right graph shows the extracted ion chromatograms (EIC) for b- and
   y-ions and the retention time for the selected proteotypic peptides in
   the 3-30 kDa fractions collected from cultural supernatants of RT ATCC
   27756 containing RUMTOR_00181. The EICs of b- and y-ions are labeled
   according to the fragmentation patterns shown in the left section. In
   the right section, observed tandem mass spectrometry (MS²) spectra are
   depicted. The illustrations detail their MS^2 with mass accuracy
   expressed in parts per million (ppm).

RORDEP-synthesizing RT strains associate inversely with adiposity

   It has been reported that the RT species has a relatively high
   prevalence and abundance in human gut microbiota^[117]19. In analyses
   of 1,493 intestinal metagenomes from three public
   datasets^[118]24–[119]26, we confirm the presence of the RT species in
   93% of the metagenomes and with a mean relative abundance of 1%
   although the latter varies from 0 to 22% at the individual level
   (Supplementary Fig. [120]5 and Table [121]5).

   We quantitated the absolute number of specific RT strains encoding
   RUMTOR_00181 and thereby releasing RORDEPs in human stools by
   integrating qPCR readouts of primers specifically targeting the RORDEP1
   sequence of RT genomes followed by flow cytometric enumeration and
   staining of faecal bacterial cell counts. In an analysis of 59 healthy
   adults (Supplementary Table [122]6), we found RT strains encoding
   RUMTOR_00181 in all examined individuals with an absolute cell count
   that varied 10^5-fold (ranging from 5.6 × 10^6 to 4.3 × 10^11 bacterial
   cells per gram of faeces) (Fig. [123]2a). In addition, in gut
   microbiota analyses of the absolute count of RT strains carrying the
   gene encoding RUMTOR_00181 from vegans and omnivores^[124]27, we find
   no significant difference between the dietary groups, suggesting that
   choice of diet may not be a major factor driving the abundance of
   RUMTOR_00181-synthesizing RT strains (Fig. [125]2b and Supplementary
   Table [126]7).

Fig. 2. RT strain abundance and plasma RORDEPs and their host correlatives.

   [127]Fig. 2
   [128]Open in a new tab

   a, Abundance of RT strains harbouring the RUMTOR_00181 gene in stools
   from healthy adults. The RORDEP1 nucleotide sequence was used as a
   marker of the RUMTOR_00181 gene. The data are normalized against the
   total bacterial cell count. The blue line indicates the median value.
   b, Abundance of RT strains carrying the RUMTOR_00181 gene in stools
   from omnivores (black) and vegans (pink), shown as jittered strip
   plots. The mean value and the standard error of the mean are overlaid
   on the individual dots. The P value was computed by Wilcoxon rank-sum
   test (unpaired, two sided). c, Inverse relationships between the
   abundance of RUMTOR_00181-positive strains of the RT species and BMI
   (left) and fat percentage (right), analysed after excluding 13
   individuals with undetectable plasma RORDEP levels. d, Association
   between BMI and the abundance of the RUMTOR_00181 gene-carrying RT
   strains in the LifeLines DEEP 2016 cohort (n = 1,135)^[129]28. The
   relationship was assessed using a linear regression model, with BMI as
   the dependent variable and the relative abundance of the RUMTOR_00181
   gene as the independent variable. β (beta) represents the regression
   coefficient from a linear regression model. Covariates include age, sex
   and sequencing read count. Only individuals with non-zero RUMTOR_00181
   abundance (orange dots) are visualized. e, Plasma RORDEP1 and RORDEP2
   peptides in 46 healthy Danish adults. Each data point represents an
   individual measurement. In the box-and-whisker plot, the box boundaries
   represent the 25th and 75th percentiles. The line in the middle denotes
   the median (50th) percentile. The whiskers span between the minimum and
   maximum values. f, Correlation between plasma concentrations of RORDEP1
   and RORDEP2. For c and f, Pearson’s two-sided correlation (ρ) tests
   were performed, and P values were adjusted for multiple comparisons
   using the Benjamini–Hochberg method in c. For c, d and f, error bands
   represent the linear regression line (black solid line) and the 95%
   confidence interval (grey shaded region).

   [130]Source data

   In human epidemiological studies, the absolute counts of
   RUMTOR_00181-synthesizing RT strains correlate inversely with BMI and
   body fat percentage (Fig. [131]2c). To further validate our initial
   observations, we analysed the LifeLines DEEP 2016 cohort^[132]28 (661
   females and 471 males, aged 45 ± 14 years) in which information on BMI
   was publicly available. Among the 1,135 individuals, we confirmed an
   inverse association between BMI and the relative abundance of the
   RUMTOR_00181 gene in the 359 metagenomes where the gene was detectable
   (β = –0.151, P = 0.002; Fig. [133]2d).

RORDEPs are present in human plasma

   We established and applied an absolute quantification (AQUA) peptide
   approach combined with SureQuant targeted proteomics ([134]Methods,
   Extended Data Fig. [135]2a and Supplementary Table [136]8), using
   isotope-labelled unique proteotypic peptides derived from each of the
   two RORDEPs as internal standards, achieving the lowest limit of
   quantification at 2.5 pM. In healthy adults (Supplementary Table
   [137]6), the plasma concentrations of RORDEP1 and RORDEP2 after an
   overnight fast show means of 176 pM and 210 pM, respectively,
   exhibiting a three- to fourfold interindividual variation (Fig.
   [138]2e). Quantification of the two peptides shows acceptable
   intra-individual coefficients of variation (Extended Data Fig.
   [139]2b), and they show a tight positive correlation (Fig. [140]2f).
   Circulating levels of RORDEPs and host BMI or body fat percentage show
   no relationship (Supplementary Fig. [141]6).

Extended Data Fig. 2. Absolute quantification of RORDEP1 and RORDEP2 peptides
in human plasma samples applying liquid chromatography tandem mass
spectrometry (LC-MS).

   [142]Extended Data Fig. 2
   [143]Open in a new tab

   In panel (a), the left section shows the selected proteotypic peptides
   (underscored and highlighted in red) that are unique to the sequences
   of RORDEP1 or RORDEP2. In the central section, the left graph displays
   the fragmentation patterns for the unique peptides in the
   identification of RORDEPs. b- and y- ions indicate that the
   fragmentation occurs in a way that the charge is retained on the
   N-terminus and C-terminus, respectively, of the peptide. The b-ions
   contain the amino-terminal end of the peptide and are numbered from the
   N-terminus (the beginning of the peptide chain) and the y-ions contain
   the carboxyl-terminal end of the peptide and are numbered from the
   C-terminus (the end of the peptide chain). The middle graph shows the
   extracted ion chromatograms for b- and y-ions, and the retention time
   for the selected proteotypic peptides in human plasma. The extracted
   ion chromatograms (EICs) of b- and y- ions are labeled according to the
   fragmentation patterns that are shown in the left graph. In the right
   section, observed tandem mass spectrometry (MS²) spectra using a Q
   Exactive mass spectrometer is depicted. The section details their MS 2
   with mass accuracy expressed in parts per million (ppm). (b) Bar graphs
   representing the coefficient of variation for intra- individual
   variability in signal intensity of RORDEP1 (blue) and RORDEP2 (orange).
   Each bar represents one individual sample out of five independently
   examined samples from the same plasma pool.

   [144]Source data

An RT strain expressing both RORDEPs improves host metabolism

   Following a feasibility study (Supplementary Fig. [145]7), we conducted
   interventions in high-fat-fed mice given oral gavage twice weekly with
   the RT ATCC 27756 strain expressing both RORDEPs, here named RT2 in a
   dose of 5 × 10^9 colony-forming units (CFU) or heat-killed RT2 in an
   equivalent amount or phosphate-buffered saline (PBS) containing 10%
   glycerol (Fig. [146]3a). At the end of the 8-week intervention, the
   engraftment was quantified by qPCR. Compared with the PBS group, the
   live-RT2-treated group showed a fivefold increase in faecal RT2
   abundance (Extended Data Fig. [147]3a), while having a minimal impact
   on the overall gut microbiota composition (Extended Data Fig. [148]4).

Fig. 3. The RT2 strain expressing both RORDEPs improves mouse metabolism.

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

   a, Lean male C57BL/6N mice were given a high-fat diet for 8 weeks along
   with twice-weekly oral gavage of live RT2 at a dose of 5 × 10^9 CFU per
   100 µl of sterile PBS containing 10% glycerol. Control groups received
   sterile PBS containing 10% glycerol or an equivalent dose of
   heat-killed RT2. b, An ipGTT was done at week 6. c, Relative body
   weight changes in mice following the intervention. Weight increase was
   normalized to baseline at week 0. d, Body weight changes between week 0
   and 8 across the three groups of mice. e, Weights of inguinal and
   epididymal white adipose tissue (measured at termination). f,
   Representative H&E-stained sections of inguinal white adipose tissue
   from each group (n = 9 images per group). The scale bar in the bottom
   of f applies to all images in each panel. g, Surface areas of segmented
   adipocytes from representative H&E-stained sections of inguinal white
   adipose tissue in f. Box plot elements: the centre line represents the
   median; the box limits indicate the 25th and 75th percentiles; the
   whiskers extend from the minimum to the maximum values. h, mRNA
   expression levels of genes related to thermogenesis and browning,
   lipogenesis, lipolysis and inflammation in inguinal white adipose
   tissue. i, 3D cross-sectional micro-CT images of femoral bone, with red
   sections indicating regions selected for cortical thickness analysis.
   j, Cortical thickness measurements derived from 3D images (n = 5 or 6
   mice per group). Data are presented as mean ± s.e.m. Unless otherwise
   stated, n = 8 or 9 mice per group. Statistical analysis used two-way
   ANOVA with Bonferroni post hoc correction for b and c, and one-way
   ANOVA with Dunnett’s post hoc correction for e, g, h and j. Panel a
   created with [151]BioRender.com.

   [152]Source data

Extended Data Fig. 3. Intestinal engraftment, body composition, brown adipose
tissue expression of uncoupling 1 (UCP1) protein, and femoral cortical
thickness in mice following eight weeks of oral gavage twice weekly of RT
ATCC 27756 strain (RT2) expressing the RUMTOR_00181 gene.

   [153]Extended Data Fig. 3
   [154]Open in a new tab

   C57BL/6N mice were eight weeks old at start of intervention and were
   given a high-fat diet for a further eight weeks. The dose of RT2 was
   5 × 10^9 in 100 µl of sterile phosphate-buffered saline (PBS) with with
   10% glycerol or heat-killed RT2. (a) Engraftment testing at end of
   intervention at week eight (n = 9 per group). Abundance of RT strains
   expressing the RUMTOR_00181 gene was measured by quantitative
   PCR-normalized 16S rDNA in faecal DNA from mice treated with PBS or the
   live RT2 strain. (b) Segmented adipocytes from representative
   hematoxylin- and eosin-stained sections of inguinal white adipose
   tissue are depicted in Fig. [155]3f and analyzed using
   AdipoCount^[156]51. (c) Magnetic resonance imaging scanning of mouse
   body composition (n = 10 for PBS and heat-killed RT2 groups, and n = 9
   for RT2 group). (d) The frequency distribution of adipocyte cell
   surface area (in arbitrary units, a.u.) in inguinal white adipose
   tissue across the three specified groups. (e) Immunoblotting result of
   UCP1- and housekeeping β-actin proteins in interscapular brown adipose
   tissue (n= six mice in each of four groups). (f) Quantifications of the
   immunoblottings of UCP1 and β-actin in panel c are given as ratios
   (n = 6 per group). (g) High-resolution 3D reconstructions of femoral
   bones (5–6 mice per group), with red sections indicating the regions
   selected for cortical thickness analysis. In panel a, significance was
   obtained by Student’s t-test (two-sided). In panels c and f,
   statistical analysis was performed using one-way ANOVA with Dunnett’s
   post hoc correction for multiple comparisons. For panels a, c, and f,
   data are presented as mean ± SEM.

   [157]Source data

Extended Data Fig. 4. Impact of R. torques ATCC 27756 (RT2) intervention on
the 16S rRNA gut microbiota profile of mice.

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

   (a) Bar plots representing the relative abundance of bacterial families
   in the gut microbiota of mice treated with RT2 (n = 9) and control mice
   receiving sterile phosphate-buffered saline (PBS) containing 10%
   glycerol (n = 9) over an 8-week period via oral gavage. (b) Box plots
   illustrating the Chao1 richness estimator and Shannon diversity index
   for the gut microbiota of RT2-treated and control mice (n = 9 per
   group). Statistical significance was calculated using the two-sided
   Wilcoxon test; ns denotes non-significant. Box plot elements: centre
   line represents the median; box limits indicate the 25th and 75th
   percentiles; whiskers extend from the minimum to the maximum values.
   (c) Principal coordinate analysis (PCoA) plot visualizing the
   Bray-Curtis dissimilarity at the amplicon sequencing variants (ASVs)
   level, comparing the gut microbial communities of RT2-treated and
   control mice. PERMANOVA analysis (two-tailed) based on Bray-Curtis
   dissimilarity was conducted to assess differences between groups.

   [160]Source data

   The RT2 treatment results in an improved glucose tolerance and a
   lowering of body weight gain, fat tissue mass and fat cell size, but an
   increase in lean body mass (Fig. [161]3b–g and Extended Data Fig.
   [162]3b–d). In parallel, we find an upregulation of uncoupling protein
   1 (UCP1) expression in the butterfly-shaped interscapular brown adipose
   tissue in the live-RT2-treated group (Extended Data Fig. [163]3e,f).
   Thermogenic gene markers, including Ucp1, Cidea and Dio2, show an
   enhanced expression in the live-RT2-treated mice, whereas expression of
   genes involved in lipogenesis, including Fasn, Scd1 and Acaca, is
   reduced in inguinal white adipose tissue. Genes involved in lipolysis
   are upregulated while the expression of a key adipose inflammation
   marker, Tnf, is diminished (Fig. [164]3h). Finally, the cortical
   thickness of femoral bone is increased in the RT2-treated mice (Fig.
   [165]3i,j and Extended Data Fig. [166]3g). By contrast, mice treated
   with an RT ATCC 35915 strain (here called the RT3 strain) lacking
   RUMTOR_00181 show no change in glucose tolerance or weight development
   but exhibit increased expression of markers of lipogenesis and
   inflammation in adipose tissue (Supplementary Fig. [167]8), probably
   attributable to the presence of virulence factors^[168]29.

An engineered E. coli strain expressing RORDEP1 improves host metabolism

   To examine whether the metabolic benefits elicited by intervention with
   the RT2 strain are specifically attributable to RUMTOR_00181 and
   thereby expression of RORDEPs independent of the R. torques genomic
   background, we engineered E. coli Nissle 1917 (EcN) to express RORDEP1
   (EcN-RORDEP1; Extended Data Fig. [169]5a–c). Diet-induced obese (DIO)
   mice on a high-fat diet received a daily oral gavage of 5 × 10^10 CFU
   of either EcN-RORDEP1 or control EcN for 4 days, with streptomycin
   added to the drinking water to promote engraftment. Consistently across
   two sets of independent interventions with monitoring periods of
   18–21 days, mice treated with EcN-RORDEP1 showed improved glucose
   tolerance and a more pronounced reduction in body weight compared with
   controls, with the weight difference persisting throughout the study
   periods (Extended Data Fig. [170]5d–m). In one of the experiments,
   engraftment levels were assessed and found to remain high, and the E.
   coli abundance in the ileum, caecum and colon was similar between
   groups (Extended Data Fig. [171]5h,i).

Extended Data Fig. 5. Mouse interventions with Escherichia coli Nissle 1917
(EcN) engineered to express RORDEP1.

   [172]Extended Data Fig. 5
   [173]Open in a new tab

   (a) Luciferase reporter assay showing the signal intensity of secreted
   RORDEP1 in the spent culture supernatant of the engineered EcN strain
   and the negative control. Analyzed at 20-fold and 200-fold dilutions.
   Culture medium alone (lank; 2 × YT) was included to assess background
   signal. Bars represent the mean luciferase signal from three biological
   replicates. (b) Quantification of secreted RORDEP1 from the engineered
   strain using a standard curve generated with a HibiT-tagged control
   protein. (c) Bacterial load of the engineered EcN strains expressed as
   colony-forming units per milliliter (CFU/mL) of diluted culture media.
   (d) In independent experiment #1, twenty-four 20-week-old C57BL/6 N
   male mice, which had been on a high-fat diet for 12 weeks prior to the
   start of interventions, were assigned to each group receiving either
   5 × 10^10 CFU of live EcN-RORDEP1 /100 µl of sterile phosphate-buffered
   saline (PBS) or 5 × 10^10 CFU of live EcN-Control /100 µl of sterile
   PBS for four consecutive days. The timeline details the protocol for
   mice acclimatization and subsequent administration of EcN strains. From
   day minus four and until termination (day 21), streptomycin was added
   to drinking water to ensure a consistently high level of bacteria
   engraftment. (e and k) Blood glucose levels measured during
   intraperitoneal glucose tolerance test (ipGTT) in two independent
   experiments. (f and l) The area under the curve (AUC) for the ipGTT in
   two independent experiments. (g and m) Graphs illustrating the body
   weight alterations over time for both groups in two independent
   experiments. (h) This plot displays the engraftment profiles in mouse
   stools of the two different EcN strains; analyses were done at day 4,
   8, 12 and 18, respectively, during a 20-day period in the independent
   experiment #1 as shown in panel a. The CFU per gram of feces are
   plotted on a logarithmic scale. Two groups are compared: the control
   group (EcN-Control, black circles) and the group treated with EcN-
   RORDEP1 (green circles). Data points represent mean CFU values, and
   error bars indicate standard deviations. (i) Distribution of E. coli
   expressed in CFU per gram of intestinal contents from various regions
   of the intestinal tract (ileum, cecum, and colon) on day 20
   (termination) in the independent experiment #1 as shown in panel (d).
   Each data point represents an individual sample from either EcN-Control
   (black) or EcN- RORDEP1 group (green), with the mean values indicated
   by the horizontal lines. (j) Overview of the second experiment lasting
   for 18 days. Statistical assessments for blood glucose levels and body
   weight changes were conducted using two-way ANOVA with Bonferroni post
   hoc adjustments. The significance of differences in AUC (panels f and
   l) and colonization (panel i) was determined with a two-sided unpaired
   Student’s t-test. Panels d and j created with [174]BioRender.com.

   [175]Source data

   Differences in animal models (lean versus DIO mice), dosage regimens
   and administration routes used across the RT and E. coli experiments
   may potentially explain the observed differences in the onset and
   magnitude of the RORDEP-induced metabolic responses.

RORDEP1 induces a rise of plasma GLP1, PYY and insulin but a decline of GIP

   To further explore the metabolic roles of RORDEPs, both were
   recombinantly produced in an E. coli culture using standard protein
   expression protocols ([176]Methods and Supplementary Fig. [177]9a,b).
   Small-angle X-ray scattering experiments show that RORDEP1 is a monomer
   ([178]Methods and Supplementary Fig. [179]9c). However, thermal
   stability tests indicate that heat exposure induces dimerization of
   r-RORDEP1 aligning with some degree of thermal instability
   (Supplementary Fig. [180]9d).

   As irisin was the lead protein in the discovery of RORDEPs, and as
   irisin is reported to bind to integrin receptors in adipocytes and
   osteocytes^[181]30, we investigated the potential interaction of
   r-RORDEP1 with various integrin receptors. However, we observed no
   binding between r-RORDEP1 and any of the tested integrins
   (Supplementary Figs. [182]10–[183]13).

   Next we assessed the potential bioactivity of r-RORDEP1 in various
   human cell cultures, including white adipocytes, osteoblasts, skeletal
   muscle cells, the enteroendocrine NCI-H716 cell line and rat INS-1
   cells. However, r-RORDEP1 in various supraphysiological concentrations
   does not per se elicit consistent effects in any of the cellular
   experiments (Extended Data Fig. [184]6).

Extended Data Fig. 6. Assessments of in vitro effects of recombinant RORDEP1
in human and mammalian cellular models applying gene expression profiling.

   [185]Extended Data Fig. 6
   [186]Open in a new tab

   In human white adipocytes, the expression of key adipogenic genes,
   including Pparγ (peroxisome proliferator-activated receptor gamma), Hsl
   (hormone-sensitive lipase), Gpat3 (glycerol-3-phosphate acyltransferase
   3), Zbtb7 (zinc finger and BTB domain-containing protein 7A), Dgat2
   (diacylglycerol O-acyltransferase 2), and Cd137 (tumor necrosis factor
   receptor superfamily member 9), was measured. In human osteoblasts,
   markers of osteogenic differentiation were assessed, including Osterix
   (transcription factor Sp7), ALP (alkaline phosphatase), CCL1(α)
   (chemokine ligand 1), SOST (sclerostin), RUNX2 (runt-related
   transcription factor 2), and COL1A1 (collagen type I alpha 1 chain). In
   human skeletal muscle cells, the expression of genes related to muscle
   differentiation and growth, such as Desmin (intermediate filament
   protein), MyoD (myogenic differentiation 1), SX1 (Sarcobox), MyoG1, and
   MyoG2 (myogenin isoforms), were evaluated. For human NCI-H716
   enteroendocrine cells, the response to r-RORDEP1 treatment did not
   result in any significant transcriptional changes in the GLP-1 release.
   Similarly, in rat INS-1 pancreatic beta cells, a model commonly used
   for studying insulin secretion, r-RORDEP1 exposure for one hour at
   varying concentrations showed no significant effects. rat INS-1 cell
   line, n = 6 biological replicates; for the remaing, n = 3 replicates
   per condition. Error bars are the SEM of 3-6 independent replicates.

   [187]Source data

   Similarly, intravenous infusion of r-RORDEP1 in rats does not result in
   any alterations of blood glucose levels over a 3-h monitoring period
   (Supplementary Fig. [188]14). The latter finding together with the lack
   of direct cellular effects made us hypothesize that RORDEPs exert their
   metabolic effects through indirect mechanisms and primarily when
   administered via the intestinal or peritoneal route. Therefore, we
   explored the potential effects of r-RORDEP1 on the release of
   intestinal and pancreatic hormones following an oral glucose load in
   lean rats (Fig. [189]4a). We find that a single intraperitoneal dose of
   r-RORDEP1 induces a decline of blood glucose compared with a placebo
   (Fig. [190]4b). Notably, r-RORDEP1 causes a 50% decline in the plasma
   GIP concentration while the plasma concentration of GLP1, PYY and
   insulin is increased (Fig. [191]4c–f). Plasma glucagon remains
   unaltered (Fig. [192]4g).

Fig. 4. Peritoneal delivery of r-RORDEP1 improves the metabolism of rats and
mice.

   [193]Fig. 4
   [194]Open in a new tab

   a, Schematic representation of the experimental design evaluating the
   effects of an oral glucose load and intraperitoneal injection of
   r-RORDEP1 on plasma concentrations of glucose, GIP, GLP1, PYY, insulin
   and glucagon in rats. Lean Sprague Dawley male rats 10 weeks of age
   were administered an oral glucose load (1 g kg^−1) over 0–2 min,
   followed by an intraperitoneal injection of r-RORDEP1 (0.8 mg kg^−1) or
   PBS as control (n = 12 per group). b–g, Levels of blood glucose (b),
   plasma GIP (c), plasma GLP1 (d), plasma PYY (e), plasma insulin (f) and
   plasma glucagon (g) measured at designated timepoints after glucose
   load and r-RORDEP1 injection. h, Experimental workflow for testing the
   effect of r-RORDEP1 on glucose tolerance in db/db mice. High-fat-fed
   db/db mice were administered daily intraperitoneal injections of
   r-RORDEP1 (2 mg kg^−1) or PBS (control) for 10 days (n = 6 or 7 per
   group). i, Results of the ipGTT. Blood glucose levels were measured at
   baseline and at designated timepoints following glucose administration.
   j, Experimental design evaluating the effect of intraperitoneal
   injections of r-RORDEP1 in mice. Lean C57BL/6 male mice 10 weeks of age
   were daily administered with an intraperitoneal injection of r-RORDEP1
   (1.0 mg kg^−1) or saline as control (n = 6 per group) for 7 days. k,
   Daily intraperitoneal injection of r-RORDEP1 or r-RORDEP2 in mice
   (n = 6 per group) for 7 days induces increased expression of genes
   involved in thermogenesis and browning of inguinal white adipose
   tissue. The mRNA levels of indicated genes were analysed by qRT-PCR.
   Data were analysed using Student’s t-test. For b–g, i and k, data are
   expressed as mean ± s.e.m. Statistical significance was determined
   using two-way ANOVA with Bonferroni’s post hoc test for b–g and i and
   one-way ANOVA with Dunnett’s post hoc test for k. Panels a, h and j
   created with [195]BioRender.com.

   [196]Source data

   To extend studies of the biological in vivo effects of r-RORDEP1 on a
   pathological condition, an intraperitoneal injection of r-RORDEP1 was
   given once daily for 10 days to high-fat-fed db/db mice. The r-RORDEP1
   intervention resulted in an improved glucose tolerance compared with a
   placebo (Fig. [197]4h,i). To confirm that these metabolic effects were
   specifically attributable to r-RORDEP1 and not due to non-specific
   effects of peptide administration, additional experiments were
   performed in lean mice comparing bioactivities of r-RORDEP1 and a
   scrambled sequence of RORDEP1 (Extended Data Fig. [198]7a–d). Following
   an oral glucose load combined with a single intraperitoneal injection
   of peptide, mice treated with r-RORDEP1 show improved glucose
   tolerance, whereas mice treated with the scrambled peptide have
   unchanged glucose tolerance compared with a placebo (Extended Data Fig.
   [199]7e,f).

Extended Data Fig. 7. Synthesis of scrambled recombinant RORDEP1 and testing
of its potential effect on blood glucose over two hours in lean male C57BL/6J
mice.

   [200]Extended Data Fig. 7
   [201]Open in a new tab

   (a) Amino acid sequences of r-RORDEP1 and scrambled r-RORDEP1. (b)
   Comparison of predicted physicochemical properties of native RORDEP1
   and scrambled RORDEP1. (c) AlphaFold2-predicted 3-dimentional structure
   of the two RORDEP1 peptides. (d) Experimental workflow of mouse
   experiment evaluating effects of the two RORDEP1 peptides on blood
   glucose in mice (n = 12 mice per group) for two hours. An initial oral
   glucose load (2 g/kg) was given in the fasting state immediately
   followed by an intraperitoneal injection of either PBS, r-RORDEP1 or
   scrambled r-RORDEP1; the two peptides were given at a dose of 1 mg/kg.
   (e) Curves of the glucose tolerance test at the indicated time points
   for the two peptides showing no effect of scrambled r-RORDEP1. (f) Area
   under the curve (AUC) analysis of glucose tolerance test (n = 12 per
   group). PBS denotes phosphate-buffered saline. Data are expressed as
   mean ± SEM. Statistical significance was determined using two-way ANOVA
   with Dunnett’s post hoc test for panel e and one-way ANOVA with
   Dunnett’s correction for panel f. Panel d created with
   [202]BioRender.com.

   [203]Source data

   To further compare the in vivo potency of r-RORDEP1 and r-RORDEP2, we
   administered each peptide in a similar dose via intraperitoneal
   injection to lean normal chow-fed mice once daily for 7 days (Fig.
   [204]4j). Compared with the PBS injection, r-RORDEP1 and RORDEP2 show
   equal potency in enhancing the expression of thermogenesis- and
   browning-related genes (Ucp1, Prdm16 and Dio2) in inguinal white
   adipose tissue (Fig. [205]4k).

RORDEP1 enhances insulin sensitivity of hepatic glucose production

   To mimic the natural intestinal release of RORDEPs in humans and to
   further investigate the role of gut luminal r-RORDEP1 in glucose
   homeostasis, we infused r-RORDEP1 into the rat duodenum under a
   euglycaemic pancreatic clamp (Fig. [206]5a). Before these experiments,
   the stability of r-RORDEP1 was assessed in in vitro experiments with
   simulated intestinal fluid (SIF) showing a calculated half-life of
   approximately 3 h (Extended Data Fig. [207]8a). Further experiments
   using a gut-on-chip model, complemented by ultra-performance liquid
   chromatography (UPLC) that was validated for r-RORDEP1 specificity,
   showed that r-RORDEP1 traverses an intestinal epithelial monolayer
   without impairing cell viability or tight junction integrity (Extended
   Data Fig. [208]8b–g). Moreover, following intraperitoneal injection of
   r-RORDEP1 in lean mice, the relative abundance of r-RORDEP1 in
   peripheral blood circulation declines to 50% within 45 min (Extended
   Data Fig. [209]8h,i).

Fig. 5. Intestinal delivery of r-RORDEP1 enhances the insulin sensitivity of
hepatic glucose output in rats.

   [210]Fig. 5
   [211]Open in a new tab

   a, Experimental schematic showing that lean male Sprague Dawley rats (8
   weeks old) were cannulated on day 1 and subjected to a 4-h euglycaemic
   pancreatic clamp on day 5. From 60 min to 240 min, the rats received
   intraduodenal infusions of either PBS or r-RORDEP1
   (60–200 pmol kg^−1 min^−1), alongside constant duodenal infusion of
   insulin and somatostatin. Glucose was infused intravenously as needed
   to maintain euglycaemia. ID, intraduodenal. IV, intravenous. b, Blood
   glucose levels (top) and cumulative glucose infusion rates (bottom)
   over time during the clamp procedure; P < 0.001 indicates significant
   differences in at least one RORDEP1-treated group compared with PBS. c,
   A scatter plot depicting the AUC for the glucose infusion rates needed
   to sustain euglycaemia in rats (n = 6 per group) receiving varying
   concentrations of r-RORDEP1 during a euglycaemic pancreatic clamp. d,
   Experimental schematic of additional intervention arms comparing ID
   PBS, IV r-RORDEP1 and ID r-RORDEP1 delivery. [3-³H]glucose was infused
   to monitor hepatic glucose production throughout the clamp procedure.
   e, Glucose infusion rates required to maintain euglycaemia during the
   clamp for the conditions shown in d. f, Bar graph showing glucose
   clearance rates across the study conditions described in d. g, Bar
   graph illustrating glucose production rates under the conditions
   outlined in d. Data are presented as mean ± s.e.m. Statistical
   significance was assessed using two-way ANOVA with Dunnett’s post hoc
   test (b), one-way ANOVA with Dunnett’s post hoc test (c and f) and
   two-sided unpaired Student’s t-test (g). For b, c and e–g, n = 6 rats
   per group.

   [212]Source data

Extended Data Fig. 8. Stability of recombinant RORDEP1 in intestinal fluid,
gut epithelial permeability of r-RORDEP1, and in vivo half-life of r-RORDEP1.

   [213]Extended Data Fig. 8
   [214]Open in a new tab

   (a) In vitro stability of r-RORDEP1 (dark blue) in simulated intestinal
   fluid (SIF) including buffer control (light blue). Absorption that were
   measured at 214 nm illustrates r-RORDEP1 degradation upon exposure to
   SIF; data were recorded at various time points. Amount of left
   r-RORDEP1 upon the exposure to SIF or buffer were expressed as
   normalized area under the curves (AUCs) relative to 0 min. (b)
   Representative UPLC chromatograms of purified r-RORDEP1 spiked into
   phosphate-buffered saline (PBS) at increasing concentrations,
   demonstrating consistent retention time and concentration-dependent
   signal. (c) Quantification of r-RORDEP1 peak areas at increasing
   concentrations, expressed in arbitrary units (a.u.). (d) Quantitative
   analysis of r-RORDEP1 in the basolateral compartment following apical
   application of r-RORDEP1 at various concentrations during an incubation
   for 12 h. A dose-dependent increase in r-RORDEP1 concentration was
   observed. (e) Assessment of Caco-2 cell viability by
   3-(4,5-dimethylthiazolyl-2)-2,5-diphenyltetrazolium bromide (MTT)
   viability assay after exposure to varying concentrations of r-RORDEP1.
   No significant impact on cell viability was observed compared to
   control. (f) Measurements of transepithelial electrical resistance
   (TEER) in Caco-2 monolayers following r-RORDEP1 exposure, indicating
   that tight junction integrity remained intact. (g) Representative
   chromatograms illustrating the presence and abundance of r-RORDEP1 in
   the basolateral compartment after transepithelial transport. The blank
   control in the chromatogram represents basolateral cell culture medium
   without r-RORDEP1 treatment, while the other chromatograms display
   either r-RORDEP1 reference standards or penetrated r-RORDEP1 detected
   in the basolateral medium. (h) Immunoblot analysis showing the presence
   6×His-tagged-RORDEP1 polypeptide in plasma samples collected from four
   chow-fed lean eight weeks-old C57BL/6N mice at 5, 15, 30, 60, 120 and
   240 min after intraperitoneal injection of r-RORDEP1 at a dose of
   1 mg/kg. The expected size of the r-RORDEP1 protein (~12.5 kilodalton
   (kDa)) is indicated on the left. The blot lanes correspond to time
   points after administration by immunoblotting assay using anti-His-Tag
   Antibody (Cell Signal Technology, #2365). The experiment was performed
   once. (i) The relative abundance of plasma RORDEP1 was quantified by
   calculating the ratio of the pixel intensity for each band to the
   intensity of the adjacent background at various time points following
   the injection of 6×His-tagged r-RORDEP1 (n = 4 per group). This
   measurement was conducted using ImageJ software. Each data point
   represents an individual measurement from one mouse. Box-and-whisker
   plots display the median, quartiles, and range of protein abundance for
   each time point, with individual data points overlaid. For panels a,
   and c-f, data are presented as mean ± SEM, and n = 3 per group.

   [215]Source data

   Intestinal administration of r-RORDEP1 in rats under a euglycaemic
   pancreatic clamp leads to a dose-dependent increase in glucose infusion
   rate. The infusion of r-RORDEP1 at the highest infusion rate of
   200 pmol kg^−1 min^−1 increases the glucose infusion rate about
   fourfold (8.1 ± 0.7 mg kg^−1 min^−1 for r-RORDEP1 compared with
   2.0 ± 0.1 mg kg^−1 min^−1 for PBS; P < 0.0001 (Fig. [216]5b,c). On the
   basis of these findings, an infusion rate of 200 pmol kg^−1 min^−1 of
   r-RORDEP1 was chosen for subsequent experiments.

   Next we administered isotopically labelled glucose intravenously to
   rats under pancreatic clamp conditions, allowing us to concurrently
   monitor changes in both hepatic glucose production and peripheral
   glucose uptake (Fig. [217]5d). Following intestinal infusion of
   r-RORDEP1, we find that the hepatic glucose production (HGP) rate
   declines by about 40%, whereas glucose uptake into peripheral tissues
   remains unaltered (Fig. [218]5e–g).

   To understand the underlying molecular mechanisms of the RORDEP-reduced
   HGP, another cohort of lean rats received intestinal infusion of
   r-RORDEP1 at 200 pmol kg^−1 min^−1 (Fig. [219]6a) for 3 h. Blood
   glucose levels decrease significantly (Fig. [220]6b) and liver
   transcriptome and proteome profiling show clear separation between
   r-RORDEP1- and control-treated rats (Fig. [221]6c and Extended Data
   Fig. [222]9a,b). Univariate statistical analysis returns 2,145
   significantly differentially expressed genes between r-RORDEP1- and
   control-treated groups, of which 966 genes are upregulated and 1,179
   are downregulated (Fig. [223]6d) that further annotate to 85 Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathways (Supplementary Fig.
   [224]15). Proteome analyses show that 379 liver proteins are
   differentially expressed between r-RORDEP1-treated and control groups
   (Extended Data Fig. [225]9c). Collectively, these transcriptome and
   proteome data, combined with phosphoproteomic and pathway analyses,
   show a downregulation of genes and proteins controlling liver
   gluconeogenesis, glycogenolysis and lipogenesis but an upregulation of
   key genes and proteins involved in insulin signalling, glycogenesis and
   glycolysis (Fig. [226]6e, Extended Data Fig. [227]10, Supplementary
   Table [228]9 and Supplementary Fig. [229]16).

Fig. 6. Intestinal delivery of r-RORDEP1 to rats changes the expression of
key liver genes involved in metabolism and insulin signalling.

   [230]Fig. 6
   [231]Open in a new tab

   a, A total of 12 8-week-old lean male Sprague Dawley rats fed standard
   chow were randomized to receive either intraduodenal infusion of
   r-RORDEP1 (200 pmol kg^−1 min^−1) (n = 6 in the RORDEP1 group) or
   sterile PBS for 3 h (n = 6 in the control group). b, r-RORDEP1
   administration resulted in a 13% reduction in blood glucose levels at
   180 min relative to the baseline at timepoint 0 min. Significance was
   determined using paired Student’s t-test. c, PCA score plot showing 12
   rat liver RNA samples from r-RORDEP1 or control infusion groups, based
   on 13,453 normalized gene expressions after variance stabilizing
   transformation. The plot illustrates all genes projected onto the first
   and second principal components. d, Volcano plot illustrating the
   results of univariate analysis of differentially expressed genes (DEGs)
   in the liver. Each point represents one gene, plotted with its
   log[2](fold change) on the x-axis and −log[10]-adjusted P value on the
   y-axis. Genes significantly upregulated (P[adj] < 0.05 and log[2](fold
   change) > 1) are shown in blue, significantly downregulated genes
   (P[adj] < 0.05 and log[2](fold change) < −1) are in orange, and
   non-significant genes or those not meeting the fold change criteria are
   in grey. e, Heatmap showing gene expression profiles (z-score of
   normalized expression levels) associated with gluconeogenesis,
   glycogenolysis, lipogenesis, glycogenesis, glycolysis and the
   insulin–PI3K–AKT signalling pathway. All genes shown are significantly
   differentially expressed between the two experimental groups. For d and
   e, the P values were adjusted using the Benjamini–Hochberg method for
   multiple corrections. Panel a created with [232]BioRender.com.

   [233]Source data

Extended Data Fig. 9. Changes of the rat liver proteome following intestinal
infusion of recombinant RORDEP1.

   [234]Extended Data Fig. 9
   [235]Open in a new tab

   (a) Twelve 8-week-old lean male Sprague-Dawley rats (six after
   r-RORDEP1 and six after infusion of phosphate buffered saline (PBS))
   were included. r-RORDEP1 was infused into duodenum with a rate of 200
   pmol/kg/min for three hours. The analysis involved profiling a total of
   7,678 proteins from the rat liver. (b) Principal Component Analysis
   (PCA) plot displays the distribution of samples based on the first and
   second principal components (PC1 and PC2), capturing 24.2 % and 17.3 %
   of the total variance, respectively. Each point represents an
   individual sample, categorized by treatment group (r-RORDEP1 or PBS).
   (c) Heatmap visualizes 379 proteins that exhibit significant
   differential expression (277 up- and 109 down regulated) between the
   control and r-RORDEP1-treated groups. Differential expression was
   determined using a threshold q-value < 0.05. Expression of the
   differential proteome was z-score transformed for data visualization.
   Panel a created with [236]BioRender.com.

   [237]Source data

Extended Data Fig. 10. Phosphoproteomics analysis of rat liver following
intestinal infusion of recombinant RORDEP1.

   [238]Extended Data Fig. 10
   [239]Open in a new tab

   (a) Twelve 8-week-old lean male Sprague-Dawley rats (six after
   r-RORDEP1 and six after infusion of phosphate buffered saline (PBS))
   were included. r-RORDEP1 was infused into duodenum with a rate of 200
   pmol/kg/min for three hours. The analysis involved profiling a total of
   4,877 phosphoproteins. (b) Principal Component Analysis (PCA) plot
   displays the distribution of samples based on the phosphoryl proteome
   in the group treated with PBS (P_Control) and the group treated with
   r-RORDEP1 (P_RORDEP1). (c) Visualization of the number of significantly
   differential (log2 fold change (log2FC) > 1 and adjusted p < 0.05)
   phosphoryl proteins between the two groups; up- and down-regulated
   features were colored with red and blue, respectively. (d) Volcano plot
   visualizing the changed phosphoryl proteins in panel b; proteins
   involved in hepatic glucose- and lipid metabolism are labeled. (e)
   Differential pathway enrichment analysis showing that phosphorylation
   of PPAR pathway was activated following in vivo exposure to r-RORDEP1.
   Up- and down-regulated phosphoryl proteins are marked with red and blue
   boxes, respectively. For panels c and d, the significance was
   determined using two-sided Student’s t-test, with post hoc correction
   using the Benjamini-Hochberg method. Panel a created with
   [240]BioRender.com.

   [241]Source data

Discussion

   The human microbiome is hypothesized to be a potentially rich but
   untapped resource for the discovery of novel compounds that may have a
   regulatory impact on human biology^[242]10,[243]12–[244]14,[245]17.
   Investigating this concept, we applied a genomic alignment approach to
   mine a public database for similarities between coding nucleotide
   sequences of 118 human ligands and their precursor proteins involved in
   metabolism and 285,952 representative reference genomes of prokaryotes.
   In some RT strains, we identified a gene that encodes the RUMTOR_00181
   protein consisting of 1,271 amino acid residues including two FN3
   domains and a signal peptide, the latter suggesting a secreted protein.
   The two FN3-domain-containing bacterial peptides, RORDEP1 and RORDEP2,
   with 24% and 25% amino acid identity, respectively, to the human
   hormone irisin, are predicted to be proteolytically cleaved and
   continuously released to the gut lumen.

   The RORDEP-synthesizing bacteria are prevalent and abundant in the
   human intestinal microbiota, and the presented human epidemiological
   studies, as well as our preclinical studies, suggest that these
   bacterial strains may impact host metabolism. We find that the
   abundance of RORDEP-synthesizing RT strains correlates inversely with
   BMI and body fat. In addition, interventions in high-fat-fed mice with
   an RT strain synthesizing both RORDEPs or in DIO mice with an E. coli
   strain, engineered to express RORDEP1, elicit an improved glucose
   tolerance, a higher bone density and a less high-fat-diet-driven weight
   gain. The impact on weight reduction is related to an enhanced adipose
   tissue expression of genes involved in thermogenesis, browning and
   lipolysis, and a diminished expression of lipogenic and inflammatory
   genes.

   Next, we focused on RORDEPs. Peritoneal administration of r-RORDEP1 in
   mice induces an improved glucose tolerance and an increase of
   thermogenesis markers while peritoneal injection in rats after an oral
   glucose load induces an increase in plasma GLP1, PYY and insulin but a
   decline in plasma GIP. On the basis of previous
   knowledge^[246]31–[247]35, the effects of r-RORDEP on the release of
   the mentioned hormones are expected to result in a diminished food
   intake and a relative protection against obesity driven by a high-fat
   diet. Furthermore, intestinal infusion of r-RORDEP1 improves hepatic
   insulin sensitivity by a marked reduction in insulin-mediated
   inhibition of liver glucose production coupled with downregulation of
   genes and proteins controlling gluconeogenesis, glycogenolysis and
   lipogenesis, but an upregulation of key genes and proteins involved in
   glycogenesis, glycolysis and hepatic insulin signalling.

   Irisin, the lead human peptide hormone in our discovery of RORDEPs,
   shares only modest amino acid residues with the two
   FN3-domain-containing bacterial polypeptides. Still, irisin and RORDEPs
   have in preclinical studies similar in vivo effects on adipose and bone
   tissues^[248]22,[249]30. However, unlike irisin, RORDEPs do not
   interact with integrin receptors.

   Our report comes with limitations. Effects of intestinal or intravenous
   delivery of r-RORDEP1 were tested only for a few hours and targeting
   rat liver and blood glucose homeostasis. Experiments with intestinal or
   parenteral infusion of r-RORDEPs for weeks are needed to explore the
   long-term effects of these peptides on multiple targets of rodents and
   non-human primates including adipose tissue, skeletal muscle, bone and
   brain. In addition, we have not explored whether RORDEPs being
   continuously secreted and released into both the intestines and the
   bloodstream primarily have their effects mediated via the enteric
   nervous system, via blood circulation or both. Furthermore, experiments
   are warranted to explore the mode of action of RORDEPs including the
   discovery of their receptor. Hypothetically, RORDEPs may exert
   permissive synergistic effects together with a variety of hormones and
   metabolites targeting organs such as enteroendocrine cells, liver, fat
   tissue, skeletal muscle, bones and brain, as well as enteric neuronal
   signalling directed at target tissues. Finally, randomized clinical
   trials are now awaited exploring the physiological role of RORDEPs in
   healthy individuals and the potential of RORDEPs in native or
   engineered forms for the prevention and treatment of common human
   metabolic disorders.

Methods

Bioinformatics analyses

   A database encompassing 118 human proteins and their precursors
   (Supplementary Table [250]1) related to human metabolism was curated
   through a manual search in the UniProt database
   ([251]https://www.uniprot.org/) using the keywords ‘human peptide
   hormones’, ‘human neuropeptides’ and ‘human cytokines’. The prokaryotic
   genome sequences downloaded from the NCBI database
   ([252]ftp://ftp.ncbi.nlm.nih.gov/blast/db) were searched for the
   protein and peptide sequences reported in Supplementary Table [253]1
   with a threshold e-value ≤ 0.05 and identity > 10% using tBLASTn
   (2.2.27+). Mapping results are provided in
   [254]https://github.com/fjw536/RORDEP. An additional follow-up
   bioinformatics alignment (Supplementary Table [255]2) was performed by
   tBLASTn to query the protein sequence of RUMTOR_00181 against 244
   available metagenomic or culturomic [Ruminococcus] torques strain
   genomes (as of 4 June 2025;
   [256]https://www.ncbi.nlm.nih.gov/datasets/genome/?taxon=33039).

   Alignment between the amino acid sequences of RORDEP1 and RORDEP2 and
   1,077 sequences of available human FN3-domain-containing proteins
   ([257]https://github.com/fjw536/RORDEP) were performed using NCBI
   BLASTp (2.2.27+), and phylogenetic trees were built by BLAST pairwise
   alignments (Supplementary Fig. [258]2; tree method, neighbourhood
   joining; maximal sequence difference, 0.8; and distance, Grishin (amino
   acid)). The high-quality protein structures of RUMTOR_00181 were
   modelled using the artificial intelligence algorithm AlphaFold2 (ref.
   ^[259]36). Following AlphaFold modelling, the predicted structures were
   visualized using Pymol2.5 software ([260]https://pymol.org/2/).
   Multiple sequence alignment was performed with Clustal Omega (available
   at [261]https://www.ebi.ac.uk/Tools/msa/clustalo/) to identify the
   positions of cysteine residues and to count both identical and
   conserved residues. Protein sequence identity is calculated by dividing
   the number of identical positions within the alignment by the total
   length of the alignment region. Potential virulence-associated factors
   in the genomes of RT strains ATCC 27756 and ATCC 35915 were identified
   through homology mapping against the core dataset of the Virulence
   Factors of Pathogenic Bacteria (VFDB)^[262]37, using NCBI BLASTn
   (version 2.12.0).

   The search for potential proteases and peptidases in the genome of RT
   ATCC 27756 was carried out using the MEROPS database^[263]38. This
   database serves as a comprehensive resource for information on
   peptidases (also known as proteases, proteinases and proteolytic
   enzymes) and their inhibitors. The classification of peptidases in the
   MEROPS database is hierarchical and based on structural
   characteristics.

Bacterial strains and culture conditions

   For use in mouse intervention studies, a RUMTOR_00181 gene-carrying RT
   strain (Holdeman and Moore 27756 with strain designation VPI B2-51
   (ATCC) and in our study protocol named RT2) was purchased from ATCC
   Bacteriology Collection ([264]https://www.atcc.org/). The bacterial
   strains were cultured under anaerobic conditions (95% N[2], 5% H[2])
   following handling instructions from ATCC. When the optical density at
   600 nm (OD[600]) of the bacterial culture reached 0.8, an aliquot of
   the culture was serially diluted and then plated onto brain heart
   infusion agar (BHI agar, Millipore, 70138) to determine the bacterial
   load. For oral gavage in mice, cultures of both strains were
   centrifuged at 6,000 g for 10 min, washed twice with PBS and
   concentrated with 20% (vol/vol) sterile glycerol anaerobically to
   5 × 10^10 CFU per 100 µl. Optionally, concentrated RT2 strains at
   5 × 10^10 CFU per 100 µl were diluted to 5 × 10^9 CFU per 100 µl,
   autoclaved at 121 °C for 15 min and aliquoted to have ‘heat-killed
   RT2’. Before oral administration to mice, live bacterial solution
   stocks were thawed and diluted to 5 × 10^9 CFU per 100 µl.

   In the genetically modified organism (GMO) study, we used engineered E.
   coli Nissle 1917 called the EcN (Mutaflor) strain, previously
   established as a chassis strain for advanced microbiome
   therapeutics^[265]39. This strain was modified to include a green
   fluorescent protein (GFP) gene at the transposon Tn7 attachment site
   and the aadK gene for streptomycin resistance, enhancing cultivation
   and engraftment in mouse models. The strain’s native pMUT1 plasmid was
   removed and replaced with a re-engineered version of pMUT1, including
   the kanR gene for strain selection and expression of RORDEP1. For the
   expression and release of RORDEP1 by the Gram-negative chassis, the
   sequence of the OmpA secretion signal was fused downstream to the start
   codon in the open reading frame of RORDEP1 (ref. ^[266]40). For
   detection and semi-quantification of the secreted RORDEP1, a
   bioluminescence signal was introduced downstream of the secretion-tag
   OmpA (ref. ^[267]41). To clone the fragments on pMUT1, the gBlocks
   containing the secretion and bioluminescence signal upstream of the
   coding sequence of RORDEP1 were flanked with a 25–30-nucleotide
   sequence (Integrated DNA Technologies) complementary to the primer’s
   tails designed to amplify the vector. For recombinant expression in the
   E. coli Nissle 1917 strain, the codon optimized gene fragments were
   cloned into the native cryptic plasmid pMUT1 under the control of
   previously characterized constitutive promoter MS8 (P[MS8])^[268]42. A
   ribosomal binding site (RBS) library was generated in silico to tune
   the expression levels of the peptides using the EMOPEC software^[269]43
   (Empirical Model and Oligos for Protein Expression Changes;
   [270]http://emopec.biosustain.dtu.dk). Four Shine–Dalgarno sequence
   variants (RBS1 to 4) were generated with different predicted expression
   levels. Thus, four different reverse primers were designed to amplify
   the backbone plasmid, each one flanked with a different RBS variant.
   PCRs for cloning were performed using the Phusion High-Fidelity PCR
   Master Mix with HF Buffer (Thermo Fisher Scientific, F531L). All PCR
   products were treated with DpnI (New England Biolabs, R0176S) and
   purified (GeneJET Gel Extraction Kit, Thermo Fisher Scientific, K0691)
   according to instructions from the vendors. Fragments were assembled
   using Gibson Assembly Master Mix (New England BioLabs, E2611L) as
   recommended by the vendor. A volume of 2 μl of the cloning mixture was
   transformed into 50 μl of NEB 10-beta competent E. coli cells (New
   England BioLabs, C3019H), by gentle mixing, followed by incubation for
   30 min on ice and heat-shock for 30 s at 42 °C and subsequently on ice
   for 5 min. A total of 950 μl of super optimal broth with catabolite
   repression medium (NEB 10-beta/Stable Outgrowth Medium, B9035S) was
   added and incubated at 37 °C for 1 h at 250 rpm. A 100-μl aliquot was
   spread on pre-warmed LB agar plates supplemented with 50 μg ml^−1
   kanamycin, which were incubated at 37 °C overnight. Colony PCR to
   confirm positive clones was performed using OneTaq 2× Master Mix with
   Standard Buffer (New England BioLabs, M0482S). The cloning was verified
   by Sanger sequencing (Eurofins Genomics Europe Shared Services). Four
   different plasmid vectors were generated for recombinant expression of
   RORDEP1.

   Plasmids were purified (NucleoSpin Plasmid EasyPure, MACHEREY-NAGEL,
   740727.250) and transformed into electrocompetent cells of the EcN
   strain. Briefly, an overnight culture of wild-type EcN was 100-fold
   diluted in LB broth media and incubated at 37 °C, shaking until cells
   reached an OD[600] of 0.4–0.6. Cells were washed (5 min, 4,000 g, 4 °C)
   with an equal volume of ice-cold 10% glycerol (Honeywell, 15523), three
   times in total, and concentrated 250× in ice-cold 10% glycerol. A 50-μl
   aliquot of EcN competent cells was mixed with 1 μl (50 ng μl^−1) of
   appropriate engineered plasmid, electroporated (MicroPulser
   Electroporator, Bio-Rad Laboratories) and recovered in 950 μl of super
   optimal broth with catabolite repression medium for at least 1 h. A
   100-μl aliquot was spread on pre-warmed LB agar plates supplemented
   with 50 μg ml^−1 kanamycin and 50 μg ml^−1 streptomycin, which were
   incubated at 37 °C overnight. Engineered strains were saved in 25%
   glycerol at −80 °C until further analysis. Whole-plasmid nanopore
   sequencing (Unveil Bio ApS) was performed for all plasmids purified by
   the engineered EcN strains.

   Growth profile analysis of all engineered strains was performed by
   kinetic studies on cell growth (temperature: 37 °C; run: 24 h;
   interval: 10 min; shake: linear (continuous); read: sfGFP (F) 485, 528
   and optical density (A) 600) using an Epotech 2 microplate reader
   (BioTek, Agilent Technologies) under anaerobic conditions. Individual
   colonies of all strains were inoculated in pre-reduced 2×YT broth
   medium supplemented with 50 μg ml^−1 streptomycin and 50 μg ml^−1
   kanamycin in a deep-well 96-well plate (Agilent Technologies, 204751)
   and incubated at 37 °C, shaking (250 rpm), anaerobically (BD GasPak EZ
   anaerobe pouch system, 260683), overnight. A Nunc Replication System
   (Thermo Fisher Scientific) was used to inoculate the strains in fresh
   pre-reduced 2×YT media with antibiotics. Growth curves were generated,
   and growth rates were estimated using Quantitative Growth Curve
   Evaluation software^[271]43 (QurvE;
   [272]https://www.qurveanalysis.com/app/qurve). All anaerobic procedures
   were performed in a Coy anaerobic chamber (Coy Laboratory Products),
   and all the reagents, media and consumables used were pre-reduced for
   at least 24 h. All experiments were performed in three biological
   replicates. The fitness cost of peptide secretion on cell growth was
   calculated by dividing the growth rate of the control strain with the
   growth rate of the respective engineered strain. The experiment was
   performed in biological triplicates.

   For the detection and semi-quantification of bioluminescent-tagged
   RORDEP1 in the spent bacterial fraction, individual colonies of the
   engineered strain and the negative control were inoculated in 2×YT
   broth medium supplemented with 50 μg ml^−1 streptomycin and 50 μg ml^−1
   kanamycin and incubated at 37 °C, shaking (250 rpm), overnight.
   Cultures were diluted 100-fold in the same fresh media and incubated at
   37 °C, shaking, overnight. The Nano-Glo HiBiT Extracellular Detection
   System (Promega, N2420) bioluminescent assay was performed on the
   bacterial spent fraction following the instructions provided by the
   vendor. Briefly, appropriate volumes of the LgBiT Protein and the
   Nano-Glo HiBiT Extracellular Substrate were diluted 100- and 50-fold,
   respectively. The detection mixture was added to the spent bacterial
   fraction in a 1:1 ratio. Luminescence (1 s integration time) was
   measured in an opaque 96-well plate after 5 min of incubation. A 20-μM
   HiBiT control protein (Promega, N3010) was used as positive control to
   generate the standard curve for semi-quantification of the secreted
   peptides in the spent media. An E. coli Nissle 1917 strain harbouring
   the empty pMUT1 vector was used as negative control. The experiment was
   performed in biological triplicates.

Preparation of bacterial supernatant filtrates and proteomics analysis

   The supernatant was collected from a 150-ml culture of RT2 and filtered
   using filters with several differing pore sizes and molecular-mass
   cutoffs. Swing-bucket centrifugation (Eppendorf 5415R centrifuge;
   Eppendorf) was applied for all the procedures. The supernatants were
   centrifuged at 7,000 g for 10 min at 4 °C and then passed through
   polyethersulfone filters (0.22 μm; Merck Millipore) to remove the
   residual bacterial cells. The derived filtrates were passed through
   100-kDa filters (Vivaspin 20 polyethersulfone ultrafiltration unit;
   Sartorius) at 6,000 g for 10 min. The 100-kDa filtrates were loaded
   onto 30-kDa filters (Sartorius) and centrifuged at 6,000 g for a
   further 10 min. The 30-kDa filtrates were then passed through 3 kDa
   (Pall Microsep Centrifugal Device; Pall) at 3,200 g for 10 min. The
   3–30-kDa filtrate (500 μl) was frozen at −80 °C until it was assayed.

   Methanol and chloroform precipitation of the crude protein extract was
   performed as follows: 100 μl of the protein extract was combined with
   400 μl of methanol. Then, 100 μl of chloroform was added, and the
   mixture was vortexed. Subsequently, 300 μl of LC–MS-grade water was
   added, followed by another round of vortexing. The mixture was then
   centrifuged at the maximum speed for 1 min, after which the methanol
   and water layer was carefully removed. Next, 300 μl of methanol was
   added, and the mixture was vortexed to dislodge the protein pellet.
   After spinning at maximum speed for 5 min, the chloroform and methanol
   layer was cautiously removed without disturbing the pellet. The pellet
   was then allowed to dry in tubes with open lids for approximately 5 min
   at room temperature. Finally, the protein precipitates were redissolved
   in 100 μl of denaturation buffer, which contained 8 M urea in a 10-mM
   Tris buffer.

   In the process of preparing for trypsin digestion and determining the
   required concentration of trypsin, several key steps were undertaken.
   Initially, the proteins in the sample were reduced using 5 mM
   dithiothreitol (DTT) for 1 h at room temperature. This was followed by
   the alkylation of the proteins with 10 mM iodoacetamide, also for an
   hour, but conducted in the dark to maintain the reactivity of
   iodoacetamide. Then the sample was first diluted to a concentration of
   less than 1 M urea using 900 µl of 20 mM ammonium bicarbonate.
   Subsequently, LC–MS-grade trypsin was added in a 1:50 weight/weight
   ratio. The sample was then left to digest overnight at 37 °C with slow
   shaking.

   After the digestion process, the sample was dried down in a SpeedVac
   (Eppendorf, Concentrator Plus) under the ‘V-HV’ mode. This drying was
   conducted at room temperature without heating for 1 h, after which the
   liquid level was checked. The process continued under the ‘V-AQ’
   setting until the sample was just dry. The protein extracts were
   resuspended by vortexing for 1 min in 0.1% formic acid (FA) before
   adding 0.1% trifluoroacetic acid (TFA) to a final volume of 25 µl.

   The concentration of peptides was determined using a NanoDrop
   spectrophotometer. Following this, eluted peptides were desalted using
   C18 Stage Tips, as previously described^[273]44. C18 Empore disk
   membranes (3M) were packed into the bottom of 200-µl pipette tips. The
   purification process involved washing the peptides with 0.1% TFA and 5%
   methanol in Milli-Q water and repeating this step three times. This
   washing was performed by centrifugation at 1,500 g using a specially
   designed three-dimensional-printed centrifugal block. For elution,
   60 µl of solution consisting of 0.1% TFA and 50% acetonitrile in
   Milli-Q water were used, and this step was repeated three times. The
   eluted peptides were subsequently dried at 60 °C in a SpeedVac
   centrifuge. The dried peptides were then reconstituted and sonicated in
   a solution of 5% acetonitrile and 0.1% FA. The peptide concentration
   was again measured using the NanoDrop, and a quantity of 500 ng of the
   purified peptides was prepared for injection into the LC–MS/MS system
   for analysis.

   Desalted and normalized protein samples were analysed on Orbitrap
   Exprolis 480 MS (Thermo Fisher Scientific) coupled to a Evosep One LC
   system (Evosep Biosystems). In the analysis of each sample, peptides
   were first loaded onto a column with 15 cm × 150 μm inner diameter (ID)
   with 1.9-μm C18 beads (PepSep) heated at 60 °C coupled to a 20-µm-ID
   electrospray emitter (Bruker Daltonics). The elution of peptides was
   performed over a standard 21-min gradient, with buffer A (0.1% (v/v) FA
   in water) and buffer B (0.1% (v/v) FA in acetonitrile), at a flow rate
   of 1 µl min^−1. The Q-Exactive instrument (Thermo Fisher Scientific)
   was operated in a data-independent acquisition (DIA) method for this
   process. Full MS spectra were acquired at a resolution of 60,000 with
   either an automatic gain control (AGC) target of 300% or a maximum
   injection time of 100 ms, covering a scan range from 300 to 1,650 m/z.
   The collection of MS2 spectra was conducted at a resolution of 30,000.
   This process used an AGC target set to 300%; alternatively, it used a
   maximum injection time capped at 100 ms. In addition, these spectra
   incorporated the use of higher-energy collisional dissociation (HCD)
   fragmentation, which was calibrated to a normalized collision energy
   level of 30%.

   The raw files were analysed using DIA-NN 1.8.1 in direct mode and
   visualized using Skyline (version 23.1, 64-bit). The spectra were
   matched against the UniProt database of RT ATCC 27756 (UP000003577).
   Cysteine carbamidomethyl was set as a static modification. The
   false-discovery rate was set to 0.01 for both proteins and peptides
   with a minimum length of six amino acids and was determined by
   searching a reverse database.

Human study participants

   For quantification of RUMTOR_00181-carrying strains in human stools and
   estimation of the abundance of RORDEP1 and 2 in human plasma,
   respectively, 59 healthy adult men were phenotypically characterized
   (Supplementary Table [274]6). The study was approved by the Ethical
   Committees of the Capital Region of Denmark (H-16021787). Height was
   measured with the participants not wearing shoes, and weight was
   measured to the nearest 0.1 kg, with the participants in underwear.
   Plasma glucose was measured using a colourimetric slide test, and
   plasma insulin was measured using immunoassay. The homeostatic model of
   insulin resistance was calculated from fasting insulin and glucose
   concentrations as reported previously^[275]45.

   For the assessment of the potential impact of dietary habits on the
   abundance of RT strains releasing RORDEPs in faecal samples, the
   absolute cell counts of RUMTOR_00181-carrying strains were measured in
   27 healthy adults eating a vegan diet and in 27 healthy adults eating
   an omnivorous diet (Supplementary Table [276]7). The study was approved
   by the Ethical Committees of the Capital Region of Denmark (H 3
   2012-145).

   All human studies were performed in accordance with the principle of
   Helsinki Declaration II, and written informed consent was obtained from
   each participant before they could enter the study.

Microbial genomic DNA extraction

   In human and mouse faecal samples, and in cultures of RT strains,
   metagenomic DNA extraction was performed using the NucleoSpin soil kit
   (Macherey-Nagel) according to the manufacturer’s instructions. A
   NanoDrop 2000 spectrophotometer (Thermo Scientific) was used to measure
   DNA purity and concentration.

Quantification of the abundance of RUMTOR_00181-carrying strains in human
faecal samples

   This protocol consists of two parts: a measurement of the faecal
   bacterial load and a quantification of RUMTOR_00181-carrying strains.
   For microbial load determination, 80–120 mg of frozen (−80 °C) stool
   aliquots were used as reported^[277]46,[278]47.

   For the quantitation of RUMTOR_00181-carrying strains in human stools,
   metagenomic DNA was extracted from 250 mg stool using the protocol
   described before^[279]47. Primer pairs targeting universal 16S rDNA or
   RORDEP1 gene sequence were applied for the detection of total amounts
   of bacteria and RUMTOR_00181-carrying strains, respectively, using qPCR
   assay conditions described in the above section (Supplementary Table
   [280]10). A standard curve ranging between 3 × 10^3 CFU and
   3 × 10^7 CFU was included in the qPCR assay by diluting the standard
   DNA from a pure culture of RT2. Absolute abundance of
   RUMTOR_00181-carrying strains for each sample was estimated by
   normalizing the relative abundance of RUMTOR_00181-carrying strains to
   the bacterial cell load of the individual faecal sample.

In silico analysis of the RUMTOR_00181 gene in the human gut metagenomes of
study participants in the LifeLines DEEP 2016 cohort

   To elucidate the prevalence and abundance of the RUMTOR_00181 gene
   within the human gut microbiome, in silico sequencing analysis was
   performed using metagenomic data from the LifeLines DEEP 2016
   cohort^[281]28. This analysis leveraged the
   curatedMetagenomicData^[282]48 repository (version 3.6.2), which
   encompasses a comprehensive dataset of 7,772,785 UniProt Reference
   Clusters (UniRef) gene families derived from the cohort’s metagenomic
   sequencing. The amino acid sequence of RUMTOR_00181 from [Ruminococcus]
   torques—also known as the FN3-domain-containing protein (NCBI accession
   number [283]WP_004845365.1)—was used as the query for screening the
   metagenomic data. Protein-level searches were conducted against the
   UniRef90 gene family database using the BLASTP algorithm (version
   2.7.1+) with stringent parameters: an e-value threshold of 1 × 10^−5
   and a minimum sequence identity of 40% over at least 90% of the query
   protein length. These criteria were selected to ensure high-confidence
   identification of homologous sequences corresponding to RUMTOR_00181
   within the diverse and complex metagenomic landscape of the human gut.
   Following the BLASTP screening, we quantified the presence and relative
   abundance of sequences homologous to RUMTOR_00181 using the curated
   gene family abundance tables provided by curatedMetagenomicData.

Preparation of human plasma samples and proteomics-based absolute
quantification of RORDEP1 and RORDEP2 concentrations

   A total of 20 µl of plasma was diluted with three volumes of buffer A
   and filtered using 0.2-µm AcroPrep filter plates (Pall). The filtered
   plasma was loaded onto a Multiple Affinity Removal Column (4.6 × 50 mm,
   Hu-14, Agilent) coupled to a 1260 Infinity II bio-inert
   high-performance liquid chromatography (HPLC; Agilent) according to the
   manufacturer’s instructions, and the flow-through fraction (500 µl)
   collected in a time-dependent manner. Subsequently, 50 µl of depleted
   plasma was mixed with 50 µl of 2× lysis buffer (100 mM Tris–HCl
   (pH = 8.5), 10 mM TCEP, 40 mM CAA), incubated for 10 min at 95 °C and
   digested with trypsin and Lys-C (0.5 µg each) at 37 °C overnight. The
   digest was inactivated by the addition of TFA (1% final concentration).

   AQUA peptide stock solutions were prepared by dissolving isotopically
   labelled RORDEP peptides (Supplementary Table [284]9) in 5%
   acetonitrile (ACN) to a final concentration of 5 pmol µl^−1. These AQUA
   peptides contain a labelled amino acid, allowing them to behave like
   natural peptides during chromatography and electrophoresis while being
   easily distinguishable by mass spectrometry. For the measurement of
   RORDEP in plasma samples, AQUA peptides were each further diluted and
   combined in a mix of 5 fmol µl^−1. Then 6 µl of the peptide digest and
   2 µl of the ACQUA peptide mix were loaded onto Evotips (Evosep)
   following the manufacturer’s instructions.

   For the absolute quantification of RORDEPs in the plasma pool of
   RORDEP-positive subjects, r-RORDEP1 and r-RORDEP2 were diluted in the
   depleted plasma digest derived from a pool of plasma RORDEP-negative
   subjects and measured alongside the negative- and positive-pooled
   plasma samples. The amounts of the spiked-in peptides were adjusted
   according to their ionization capacities: to 3 µl of peptide digest,
   20 fmol of peptides 1, 2 and 5–8; 40 fmol of peptide 4; and 400 fmol of
   reduced and alkylated peptide 3 were added (Supplementary Table
   [285]9).

   Peptides were then separated on a Pepsep 15-cm, 150-μM-ID column packed
   with C18 beads (1.5 μm) using an Evosep ONE HPLC system applying the
   default 30 samples per day method. The column temperature was
   maintained at 50 °C. Upon elution, peptides were injected via a
   NanoSpray source and 30 μm stainless steel emitter (Evosep) into an
   Orbitrap Ascend Tribrid mass spectrometer (Thermo) operated in
   SureQuant mode. Briefly, starting from the ‘Users Custom Panel Sure
   Quant’ template, built in Thermo Excalibur software (v.4.7), spray
   voltage was set to 2,200 V, MS data were collected over a 400–1,200 m/z
   range (120,000 resolution, 251 ms maximum injection time and AGC target
   250%). MS/MS scans were triggered by a list of heavy labelled peptides
   of interest (Supplementary Table [286]1) and acquired using 15,000
   resolution, 27 ms maximum injection time, AGC target 1,000%, 1.4 Th
   isolation window and 30 normalized collision energy. MS/MS spectra
   having a minimum of n = 4 expected target ion fragments (Supplementary
   Table [287]9), defined by unique peptide sequences that do not share
   more than a minimal degree of homology (typically less than 30% over
   any seven-amino acid stretch) with other proteins, were triggering a
   second MS/MS scan for the endogenous light peptide. Isolation offset
   was set to −4 Th or −2.67 Th, respectively, for a double- or
   triple-charge precursor. Other settings include Orbitrap resolution
   60,000, 123 ms maximum injection time, AGC target 1,000% and HCD
   collision energy 30. The total fixed cycle time, switching between all
   3 MS scan types, was set to 3 s.

   A spectral library was generated with MaxQuant (v1.6.15.0). Peak lists
   were searched against the human UniProt FASTA database combined with
   RORDEP 1–2 sequences and 262 common contaminants by the integrated
   Andromeda search engine. The false discovery rate was set to 1% for
   both peptides (minimum length of seven amino acids) and proteins.
   Carbamidomethylation of cysteine was specified as fixed modification,
   heavy lysine (+8 Da), oxidation of methionine and acetylation at the
   protein N-terminus as variable modifications. ‘Match between runs’ was
   disabled.

   SureQuant raw files were analysed using Skyline 23.1.0.268 software
   following developer tutorials and guidelines. Briefly, in ‘Peptide
   Settings’, heavy isotope modifications for lysine (+8 Da) were
   activated, while carboamidomethyl (C) was set as structural
   modifications. In ‘Transition Settings’, ion charges were set to +1 and
   +2 and ion types to ‘p, y, b’. Product ion selection was set from m/z
   >precursor to last ion-1. Ion match tolerance was 0.1 m/z, and up to
   eight product ions were allowed to be selected from the spectral
   library. The mass analyser was set to ‘Orbitrap’ and resolving power to
   60.000 for MS1 filtering. In MS/MS filtering options, acquisition mode
   was set to ‘SureQuant’ and mass analyser to ‘Centroided’. Only scans
   within 5 min of spectral library MS/MS retention time ID were selected.
   After an attentive visual inspection of the correct Skyline peptides
   integration, a matrix containing the light and heavy total area of
   fragments for each target peptide was exported and used for
   quantification. Peptide fragments used for quantification are listed in
   Supplementary Table [288]9.

Mouse gut microbial DNA extraction, 16S rRNA gene sequencing and
bioinformatics analysis

   Microbial DNA was extracted from approximately 250 mg of mouse stool
   samples using the NucleoSpin Soil Mini Kit (Macherey-Nagel) following
   the manufacturer’s protocol. The extraction process began with the
   lysis of stool samples through a combination of chemical treatment and
   mechanical disruption, which involved adding lysis buffer to zirconium
   bead tubes containing the samples and performing bead beating with a
   specialized vortex adaptor. The crude lysate was subsequently treated
   to eliminate inhibitors, ensuring thorough sample purification. The
   purified lysate was then combined with a DNA-binding solution and
   passed through a silica-based spin filter. After a two-step washing
   procedure, the DNA bound to the membrane was eluted using 10 mM Tris
   buffer. The concentration of the extracted DNA was quantified using a
   Qubit Flex Fluorometer with the Qubit dsDNA HS Assay Kit (Thermo Fisher
   Scientific).

   For sequencing, DNA libraries were constructed using the Quick-16S Plus
   NGS Library Prep Kit (Zymo Research). The V3 and V4 regions of the 16S
   rRNA gene were amplified via PCR using primers 341F (a mixture of
   CCTACGGGDGGCWGCAG and CCTAYGGGGYGCWGCAG) and 806R
   (GACTACNVGGGTMTCTAATCC). Dual eight-nucleotide indices compatible with
   Illumina sequencing were added in a subsequent PCR step. Following
   quantification with the Qubit fluorometer, the DNA libraries were
   circularized using the Element Adept Library Compatibility Workflow and
   sequenced on the Element AVITI platform using the AVITI 2 × 150
   Cloudbreak Sequencing Kit.

   Bioinformatics analysis commenced with the removal of primer sequences
   from the raw sequencing data using Cutadapt. Quality trimming was
   performed with DADA2, in which sequences were filtered based on a
   predefined Phred quality score threshold, and both forward and reverse
   reads were trimmed to uniform lengths according to their quality
   scores, retaining higher-quality reads for longer lengths. DADA2 used a
   machine learning-based parametric error model to estimate error rates
   for the reads, operating under the assumption that true sequences are
   more abundant than erroneous variants. The core sample inference
   algorithm was then applied to the filtered data, followed by the
   merging of paired-end reads that shared at least 12 bases of identical
   overlap. Chimeric sequences, identified as combinations of more
   abundant parent sequences, were removed. The cleaned and trimmed reads
   were subsequently clustered into amplicon sequence variants (ASVs) with
   97–99% similarity.

   Taxonomic classification of the ASVs was conducted using the naive
   Bayesian classifier of DADA2 in conjunction with the RDP trainset 18
   database. To assess differences in gut microbial composition between
   baseline and follow-up samples, taxonomic abundances at the family
   level were visualized using stacked bar plots. Alpha diversity was
   evaluated using the Shannon index, and statistical comparisons were
   made using a two-sided Wilcoxon test. All data analyses were performed
   using R (version 4.3.2).

Expression of 6×His-tagged RORDEP1 or RORDEP2 in E. coli

   Target DNA sequences of RORDEPs were codon optimized for E. coli strain
   BL21 (DE3). The synthesized DNA sequences were cloned into the pET-30a
   (+) vector for protein expression. The vector products were transformed
   into the E. coli strain BL21 Star (DE3). The E. coli containing RORDEPs
   in plasmids were cultured on an agar plate supplemented with ampicillin
   (50 µg ml^−1) at 37 °C overnight. A single colony was then inoculated
   into terrific broth medium containing the same antibiotic and incubated
   at 37 °C with shaking at 200 rpm. When OD[600] reached 0.8, the
   bacterial culture was induced with
   isopropyl-β-d-1-thiogalactopyranoside (IPTG) at 37 °C for 4 h. Bacteria
   were collected by centrifugation and the resulting pellets resuspended
   in lysis buffer. After sonication, the supernatants containing the
   target protein were separated by centrifugation and purified using a
   one-step purification on a Ni column (GenScript Biotech). The final
   products were kept in 50 mM Tris–HCl, 150 mM NaCl and 10% glycerol, pH
   8.0, followed by sterilization using a 0.22-μm filter, before being
   stored in aliquots. The concentration was determined using a
   bicinchoninic acid (BCA) protein assay. Protein purity and molecular
   weight were assessed using standard SDS–PAGE and intact mass
   spectrometry. An endotoxin test was performed using the ToxinSensor
   Chromogenic LAL Endotoxin Assay Kit (L00350C, GenScript Biotech), using
   a modified Limulus Amebocyte Lysate and a synthetic-colour-producing
   substrate to quantitatively detect endotoxin chromogenically in a broad
   range (0.01–1 EU ml^−1). The endotoxin levels of recombinant RORDEPs
   were <0.1 EU ml^−1.

Small-angle X-ray scattering analysis of r-RORDEP1

   The r-RORDEP1 protein was dialysed against a 20-mM Tris buffer at pH
   8.0, using Slide-A-Lyzer dialysis cassettes (G2, 3.5-kDa molecular
   weight cutoff). The protein solution was concentrated using Amicon
   Ultra-0.5 centrifugal filter units with a 3.5-kDa cutoff, to achieve
   final concentrations of 1.0, 2.0 and 6.6 mg ml^−1, respectively, for
   subsequent SAXS experiments.

   SAXS data collection was conducted at the BM29 BIOSAXS beamline at the
   European Synchrotron Radiation Facility, on 15 April 2021. The
   instrument was equipped with a Pilatus 2M detector, set at 2.83 m from
   the sample, to cover a q-range of 0.044–5.205 nm^−^1. The X-ray
   wavelength was fixed at 0.998 nm. A 500 × 100 mm^2 beam was used for
   all measurements. The samples were irradiated in a quartz glass
   capillary at 20 °C, and data were collected in 10-s exposures to
   minimize radiation damage, with frame-by-frame comparison for
   monitoring.

   Two modes of SAXS analysis were used: batch mode and a flow cell that
   was connected to a size-exclusion column (SEC-SAXS) mode. For batch
   mode, protein concentrations of 1.0 g l^−1 and 2.0 g l^−1 were used. In
   SEC-SAXS, RORDEP1 was injected at a concentration of 6.6 mg ml^−1 into
   an AdvanceBio SEC 130-Å column pre-equilibrated with buffer and run at
   a flow rate of 0.16 ml min^−1. Absolute scaling was achieved by
   comparing with scattering from sterile water, and normalization was
   based on transmitted intensity by beam-stop counter.

   SAXS data reduction was performed using PRIMUSqt from the ATSAS package
   (2.8.32). Basic analyses to determine parameters such as Guinier, p(r),
   and Vp were also conducted within the same software environment.
   Molecular graphics were generated using PyMOL (version 1.8.2.3,
   Schrödinger), and figures were produced using the Spyder software
   within the Anaconda 3 environment. Theoretical scattering curves for a
   monomeric and dimeric human irisin structure (PDB entry: [289]4LSD),
   which shares 29% sequence identity with RORDEP1, were calculated using
   CRYSOL. Fits to the experimental data allowed for the determination of
   the oligomeric state of RORDEP1 in solution. The radius of gyration
   (R[g]) and the maximum particle dimension (D[max]) were calculated from
   the SAXS data. Molecular weights were estimated using the pair
   distribution function (MW[p(r)]) and Porod’s law (MW[porod]). For batch
   SAXS, R[g] values were determined as 1.53 nm and 1.51 nm for the
   1.1 g l^−1 and 2.0 g l^−1 concentrations, respectively, both with a
   D[max] of 5.0 nm. Molecular weight estimates were 12.1 kDa and 11.7 kDa
   (MW[p(r)]), and 7.8 kDa and 8.3 kDa (MW[porod]). For SEC-SAXS, at a
   protein concentration of 6.6 g l^−1, the R[g] was measured at 1.50 nm,
   D[max] at 5.0 nm and MW[porod] at 8.5 kDa.

Thermostability characterization of r-RORDEP1 by SEC-MALS and nano
differential scanning fluorimetry analysis

   The oligomeric state of r-RORDEP1 after heat treatment was analysed
   using SEC with multi-angle static light scattering (SEC-MALS). The
   analysis was performed using an Agilent system equipped with a Wyatt
   Optilab µT-rEX refractive index detector and a Wyatt µDAWN TREOS MALS
   detector. Data acquisition and analysis were conducted with Astra
   software version 7.3.2.19. Each run used a 20-µl sample injection
   volume, with a protein load of 15–20 µg. Separation was achieved using
   an AdvancedBio SEC 300A column (2.7 µm, 4.6 × 150 mm), with 20 mM Hepes
   and 150 mM NaCl as the mobile phase. Nano differential scanning
   fluorimetry was conducted on a Panta system (NanoTemper) at a
   concentration of 3 mg ml^−1 with a temperature ramp from 20 °C to 95 °C
   with a gradient of 1.5 °C min^−1. Unfolding was detected by monitoring
   the fluorescence at 330 nm, 350 nm and the 350:330 ratio.

Cell culture experiments

   Human white preadipocytes (PromCell, C-12732, lot number 456Z005.1,
   isolated from human omentum) were cultured until 80% confluent in
   preadipocyte growth medium (PromCell, C-27410, basal medium
   supplemented with 0.05 ml ml^−1 fetal calf serum, 0.004 ml ml^−1
   endothelial cell growth supplement, 10 ng ml^−1 epidermal growth
   factor, 1 μg ml^−1 hydrocortisone, 90 μg ml^−1 heparin) and switched to
   preadipocyte differentiation medium (PromCell, C-27436, basal medium
   supplemented with 8 μg ml^−1 d-biotin, 0.5 μg ml^−1 insulin,
   400 ng ml^−1 dexamethasone, 44 μg ml^−1 3-isobutyl-1-methylxanthine
   (IBMX), 9 ng ml^−1 l-thyroxine, 3 μg ml^−1 ciglitazone). The
   differentiation process to mature adipocytes was considered completed
   after 12–14 days. The treatment of r-RORDEP1 and PBS started from the
   third day of differentiation to the end with refreshment at every other
   day. Cells were collected after 14 days of differentiation, and
   thermogenic and lipogenic genes were quantified by qPCR.

   Human osteoblasts (Promocell, C-12720, lot number 445Z012.2, isolated
   from human femoral heads) were cultured in growth media (Promocell,
   C-27001, basal medium supplemented with 10% fetal calf serum) until 80%
   confluent and switched into differentiation media containing
   differentiation-promoting supplements of 10 mM β-glycerophosphate and
   50 μg ml^−1 ascorbic acid in α-MEM (Fisher Scientific, 15430584). On
   days 3, 6 and 9 of differentiation, cells were treated with r-RORDEP1
   and PBS in the presence of β-glycerophosphate and ascorbic acid. Cells
   were collected on day 10. Transcription regulators and osteoblast
   differentiation marker genes were quantified by qPCR.

   Human skeletal myoblasts (PromoCell, C-12530, lot number 451Z031.15,
   isolated from human musculus pectoralis major) were cultured in growth
   media (PromoCell, C-23060, basal medium supplemented with 5% fetal calf
   serum, 50 μg ml^−1 fetuin, 10 ng ml^−1 epidermal growth factor,
   1 ng ml^−1 basic fibroblast growth factor, 10 μg ml^−1 insulin,
   0.4 μg ml^−1 dexamethasone) until 80% confluent and switched into
   differentiation media (PromoCell, C-23061, basal medium supplemented
   with 10 μg ml^−1 insulin) developed to induce the fusion of myoblasts
   to form myotubes with typical multinucleated syncytia. After 4 days,
   extensive formation of multinucleated syncytia was observed. For a
   stable differentiation of skeletal muscle cells, after 5 days of
   incubation in differentiation medium, cells were switched back to
   growth medium. Treatment of r-RORDEP1 and PBS started from the third
   day of differentiation. Cells were collected after 10 days of
   differentiation and myogenesis genes were quantified by qPCR.

   GLP1-secreting human cell line NCI-H716 (ATCC, CCL-251) cells were
   seeded at a concentration of 1 × 10^5 cells per well onto a
   poly-d-lysin-coated plate (Sigma-Aldrich, P8920). The cells were
   cultured in RPMI 1640 medium (Thermo Fisher, 11875093) supplemented
   with 10% (v/v) heat-inactivated newborn calf serum (Thermo Fisher,
   16010167), with the medium being refreshed every other day until the
   cells reached 80–85% confluence. After reaching confluence, the cells
   were incubated for 2 h in KRH buffer for baseline secretion or KRH
   buffer supplemented with either (1) 10 mM glucose and 10 µM forskolin
   (Sigma-Aldrich, F6886)/IBMX (Sigma-Aldrich, I5879) as positive
   stimulants or (2) r-RORDEP1 at concentrations of 1.5 nM, 15 nM and
   150 nM. The KRH buffer was made in house, consisting of 138 mM NaCl,
   4.5 mM KCl, 4.2 mM NaHCO[3], 1.2 mM NaH[2]PO[4], 2.5 mM CaCl[2], 1.2 mM
   MgCl[2] and 10 mM HEPES, supplemented with 0.1% (wt/vol)
   fatty-acid-free BSA (Sigma-Aldrich, catalogue number A-603-10G),
   adjusted to pH 7. Supernatants were collected and centrifuged at
   1,500 g for 5 min at 4 °C to remove any floating cells or debris. GLP1
   levels in the supernatants were then quantified using a GLP1 EIA Kit
   (Sigma-Aldrich, RAB0201), following the manufacturer’s instructions.

   Rat insulin-secreting INS-1E 832/13 cells (a gift from B. Emanuelli,
   Novo Nordisk Foundation Center for Basic Metabolic Research, Faculty of
   Health and Medical Sciences, University of Copenhagen) were grown in a
   monolayer at 37 °C in a humidified incubator gassed with 5% CO[2], in
   RPMI 1640 medium containing 2 mM glutamine supplemented with 10%
   heat-inactivated fetal bovine serum, 100 IU ml^−1 penicillin,
   100 μg ml^−1 streptomycin, nonessential amino acids (all from Life
   Technologies, Invitrogen), 10 mM HEPES (pH 7.4), 1 mM sodium pyruvate
   and 50 μM of 2-mercaptoethanol. At 70% confluence, the cells were
   treated with r-RORDEP1 and PBS for 24 h. After the treatments, the
   supernatant media were collected for enzyme-linked immunosorbent assay
   (ELISA) measurement (Sigma-Aldrich, RAB0904) of secreted insulin levels
   at 1 h and 24 h.

   In all listed cell culture experiments, r-RORDEP1 was added at
   concentrations listed in Extended Data Fig. [290]6.

Assessment of the binding between r-RORDEP1 and integrin receptors using SPR
assay

   In the surface plasmon resonance (SPR) assay on a Biacore 8K instrument
   (Cytiva) assessing the binding interactions between Fc-tagged RORDEP1
   (Fc-RORDEP1) and various human integrin receptors, Fc-RORDEP1 was
   diluted to 10 μg ml^−1 in running buffer containing 10 mM HEPES, 150 mM
   NaCl, 3 mM EDTA and 0.005% Tween-20 at pH 7.4. Each of the 24 human
   integrin subunits was prepared at 1,000 nM in the same buffer. At
   25 °C, Fc-RORDEP1 was captured onto a Series S Sensor Chip Protein A by
   injecting it at 10 μl min^−1 until approximately 1,000 response units
   (RU) were achieved. Integrin analytes were then injected over the
   immobilized Fc-RORDEP1 at 1,000 nM, with an association phase of 60 s
   and a dissociation phase of 150 s at a flow rate of 30 μl min^−1. The
   sensor surface was regenerated after each cycle using 10 mM glycine–HCl
   (pH 1.5) at 20 μl min^−1 for 30 s. Binding interactions were monitored
   in real time and processed by referencing against a blank flow cell to
   correct for non-specific binding. Interactions were evaluated
   qualitatively based on response units during the association and
   dissociation phases. Negative controls included injections of running
   buffer over the Fc-RORDEP1 surface and integrin analytes over a
   reference surface without immobilized Fc-RORDEP1. All experiments were
   conducted in duplicate.

   In the SPR assay assessing affinity between r-His-RORDEP1, echistatin
   TFA and biotinylated human ITGAV&ITGB5 heterodimer protein in the
   presence of three running buffer solutions, the biotinylated integrin
   αVβ5 heterodimer was immobilized onto a streptavidin (SA) sensor chip
   by injecting 20 μg ml^−1 of the protein in running buffer at
   10 μl min^−1 until approximately 4,000 RU were achieved. Four running
   buffers were prepared: buffer A: 1× HBS-N (10 mM HEPES, 150 mM NaCl)
   and 0.05% Tween-20, pH 7.4, supplemented with 1 mM CaCl[2] and 1 mM
   MgCl[2]; buffer B: buffer A with 1 nM Hsp90aa1; buffer C: buffer A with
   1 mM MnCl[2]; and buffer D: buffer B with 1 mM MnCl[2]. Analytes
   (r-His-RORDEP1 and echistatin TFA) were prepared in each running buffer
   immediately before injection. The analytes were injected over the
   immobilized integrin at a flow rate of 30 μl min^−1, with an
   association phase of 60 s and a dissociation phase of 150 s. The sensor
   surface was regenerated between runs as needed using appropriate
   regeneration conditions. Binding interactions were monitored in real
   time and processed using the Biacore control software, referencing
   against a blank flow cell to correct for non-specific binding.
   Responses were evaluated based on the magnitude of RU during
   association and dissociation phases. All experiments were performed in
   duplicate to ensure reproducibility.

ELISA-based evaluation of binding affinity between r-RORDEP1 and integrin
heterodimers

   An ELISA-based assay was conducted to evaluate the binding affinity
   between RORDEP1 and various integrin heterodimers. In the first setup,
   ELISA plates were coated with recombinant His-tagged RORDEP1 or
   His-tagged fibronectin at a concentration of 5 µg ml^−1 (100 µl per
   well) and incubated overnight at 4 °C. Gradient dilutions
   (0.001–10 µg ml^−1) of integrin heterodimers (Acro
   Biosystems)—including αVβ1 (IT1-H82W6), α4β1 (IT1-H82W1), α5β1
   (IT1-H82Wa), α8β1 (IT1-H82Wb), αVβ3 (IT3-H82W9), αVβ5 (IT5-H82Wa), αVβ6
   (IT6-H82E4) and αVβ8 (IT8-H82W5)—were added to the wells and incubated
   for 2 h at room temperature. Bound integrins were detected using a
   streptavidin–horseradish peroxidase conjugate (Thermo Fisher, 21126)
   diluted 1:10,000. In the second setup, ELISA plates were coated with
   the integrin heterodimers at 5 µg ml^−1, and gradient dilutions of
   Fc-tagged RORDEP1 (0.001–10 µg ml^−1) were added to the wells.
   Detection was performed using an anti-human IgG, Fcγ fragment-specific
   antibody diluted 1:10,000. After incubation with detection antibodies
   in both assays, 3,3′,5,5′-tetramethylbenzidine substrate was added for
   colour development, and the reaction was stopped with 2 N sulfuric
   acid. Absorbance was measured at 450 nm, and binding affinities were
   determined by plotting absorbance values against protein concentrations
   to generate binding curves and calculate EC[50] values.

Fluorescence polarization assay for recombinant RORDEP1–integrin binding

   To assess the binding of r-RORDEP1 to integrins, a fluorescence
   polarization (FPOL) assay was used. r-RORDEP1 was first labelled with
   either TAMRA-NHS ester or sulfo-Cy5-NHS ester (Lumiprobe). For this,
   100 µl of r-RORDEP1 at a concentration of 130 µM was incubated with a
   1.5× molar excess of the respective dye for 4 h at room temperature,
   targeting preferential labelling at the N-terminus. The labelled
   protein was then subjected to buffer exchange with 20 mM HEPES and
   150 mM NaCl, pH 7.4, to remove unreacted dye. The final protein
   concentration was 70 µM, with a degree of labelling estimated at 0.6.

   The FPOL assay was performed with an r-RORDEP1 tracer concentration of
   70 nM in a buffer containing 20 mM HEPES, 150 mM NaCl, 0.05% Tween-20
   and 2 mM MnCl[2], at pH 7.4. Integrin αVβ3 was used at a concentration
   of 340 nM, and integrin αVβ5 at 500 nM. Samples were incubated for up
   to 3 h at 22 °C, making sure that equilibrium was attained. FPOL was
   measured on a Tecan Spark fluorescence plate reader in a 384-well plate
   (Greiner 784900). For detecting TAMRA fluorescence, a 535/25 nm
   excitation filter and a 595/35 emission filter were used; for detecting
   Cy5 fluorescence, a 620/20 nm excitation filter and a 680/30 nm
   emission filter were used. The data were not corrected for different
   detection efficiencies of the parallel and perpendicular detectors
   (G-factor 1), which is justified as the focus is on relative changes.

Stability assessment of r- RORDEP1 in simulated intestinal fluid

   SIF was prepared as recommended^[291]49. In brief, 25 mg of pancreatin
   (8× United States Pharmacopeia activity) was dissolved in 20 ml of
   KH[2]PO[4] (6.8 mg ml^−1, pH 6.8) to afford 1× UPS activity SIF. The
   resulting mixture was vortexed for 1 min, followed by sonication for
   15 min at 25 °C. The final suspension (pancreatin does not dissolve
   fully) was centrifuged and syringe filtered before use.

   A total of 520 μl of freshly prepared SIF was preincubated at 37 °C for
   15 min. The r-RORDEP1 stock solution (80 μl of 375 μM) was added to the
   fluid, and the mixture was gently stirred by pipetting and incubated at
   37 °C (600 μl total volume; 30 μM final r-RORDEP1 concentration).
   Control experiments were performed in parallel in a KH[2]PO[4] buffer
   (520 μl) by spiking it with 80 μl of RORDEP1 and incubating at 37 °C
   (30 μM final r-RORDEP1 concentration). A total of 75 μl of sample was
   drawn at 0, 15, 30, 60, 120, 240 and 480 min and quenched by ice-cold
   5% TFA in H[2]O solution (75 μl). All samples were centrifuged,
   filtered and analysed with UPLC (to determine the amount of ligand
   left) and LC–MS (to identify cleavage products of r-RORDEP1).

Transport of r-RORDEP1 across a monolayer of epithelial cells using a
gut-on-chip model

   Caco-2 cells (ATCC, HTB-37) were cultured in Dulbecco’s modified Eagle
   medium (DMEM, Gibco) supplemented with 20% fetal bovine serum (Gibco)
   and 1% penicillin–streptomycin (Gibco) to prepare the Caco-2 culture
   medium. The cells were maintained in T25 culture flasks within a
   humidified incubator at 37 °C with 5% CO[2], and 5 mL of the culture
   medium was replaced every 2 days. Once the cells reached over 85%
   confluence, they were passaged at a 1:2 ratio using standard
   trypsinization techniques.

   For the construction of the gut-on-chip model, high-throughput membrane
   barrier chips (Avatarget) were transferred into a laminar flow hood to
   ensure sterility. Caco-2 cells at ≥85% confluence was detached using
   1 ml trypsin (Gibco) and resuspended in 1 ml of Caco-2 culture medium,
   followed by cell counting. A suspension containing 2 × 10^5 cells per
   cm^2 was prepared by adding 200 μl of the resuspended Caco-2 cells to
   the upper channel of each chip. The chips were then placed in a cell
   culture incubator at 37 °C with 5% CO[2] and allowed to adhere for 2 h.
   After the adhesion period, microscopic examination confirmed that the
   majority of Caco-2 cells had firmly attached to the membrane.
   Subsequently, 600 μl of Caco-2 culture medium was added to both the
   upper and lower channels of each chip, and the chips were returned to
   the incubator. The culture medium was refreshed every 48 h. After
   7 days of incubation, the integrity of the epithelial monolayer was
   verified by measuring the transepithelial and transendothelial
   electrical resistance (TEER)^[292]50, with values ≥ 500 Ω cm^−^2
   indicating successful construction of the gut-on-chip model, suitable
   for subsequent peptide permeability assays.

   TEER measurements were conducted using a WPI EVOM2 STX2 meter. The
   chopstick electrodes were first equilibrated in PBS for 15 min until
   the instrument stabilized at 0 Ω. The electrodes were then carefully
   inserted into the designated detection wells of the membrane barrier
   chip and the upper channel, ensuring that both electrodes were
   submerged in the culture medium. Once the readings stabilized, the TEER
   value was recorded and calculated using the formula TEER = T (Ω) × A
   (cm^2), where T represents the measured resistance and A is the area of
   the model.

   For the peptide permeability assay, r-RORDEP1 was prepared in DMEM at
   final concentrations of 75 μM, 150 μM and 300 μM. FITC-Dextran 4000
   (Biotiny) at 1 mg ml^−1 was used as a positive control. The
   experimental setup included four groups: 75 μM for 12 h, 150 μM for
   12 h, 300 μM for 12 h and a positive control with FITC-Dextran 4000 for
   12 h, each with triplicate wells. For each experiment, 200 μl of the
   respective peptide or positive control solution was added to the upper
   channel, and 500 μl of DMEM culture medium was added to the lower
   channel. The chips were incubated under the specified conditions, and
   after the incubation periods of 12 h, the culture medium from the lower
   channel was collected and stored at −80 °C for subsequent
   chromatographic analysis.

   Cell viability was assessed using the CellTiter-Glo 2.0 Assay
   (Promega). The assay reagent was equilibrated to room temperature and
   mixed with Caco-2 culture medium in a 1:1 ratio. After aspirating the
   culture medium from both channels of the chip, 200 μl of the mixed
   reagent was added to the upper channel. The chip was then placed on a
   shaker plate and gently agitated for 5 min to ensure thorough mixing,
   followed by a 20-min incubation at room temperature. Subsequently,
   100 μl of the luminescent solution from the upper channel was
   transferred to a 96-well plate with transparent bottoms (Corning) and
   measured using a Tecan PRO M200 microplate reader. Cell viability was
   expressed as a percentage relative to control samples, calculated by
   comparing the OD values.

   Chromatographic detection of peptide samples was performed using an
   HPLC system (Agilent 1260) equipped with a NanoChrom BioCore SEC-150
   column (7.8 × 300 mm, 5 µm). The mobile phase consisted of 50 mM
   phosphate buffer and 300 mM sodium chloride in 10% acetonitrile, with
   isocratic elution at a flow rate of 1 ml min^−1. The column temperature
   was maintained at 30 °C, and detection was carried out using a variable
   wavelength absorbance detector set to 209 nm. Samples were injected in
   volumes of 10 μl using an autosampler. In addition, analysis of the
   positive control was conducted using an ultra-high-performance liquid
   chromatography system (Agilent 1290) with the same type of column. The
   mobile phase and column temperature were identical to the peptide
   analysis, but the flow rate was adjusted to 0.7 ml min^−1. Fluorescence
   detection was performed with an excitation wavelength of 492 nm and an
   emission wavelength of 520 nm, and samples were similarly injected at
   10 μl using an autosampler. To validate the specificity of the UPLC
   detection peak corresponding to r-RORDEP1, PBS alone was analysed as a
   negative control, followed by analysis of PBS spiked with increasing
   concentrations of purified r-RORDEP1. Dose-dependent changes in peak
   intensity and consistency in retention times were monitored to confirm
   signal specificity.

Experiments in mice

   Male mice with a C57BL/6 background (specific pathogen-free grade) and
   with varying ages as detailed below were purchased from Janvier Labs.
   All experiments in mice were conducted in accordance with approved
   protocols. The intervention study involving the RT strain and the GMO1
   study were approved by the Danish Animal Experiments Inspectorate
   (license numbers 2018-15-0201-01491 and 2020-15-0201-00568,
   respectively) and by the University of Copenhagen (project numbers
   P20-392 and P23-145, respectively). The comparative study assessing the
   effects of r-RORDEP1 and scrambled r-RORDEP1 on glucose tolerance,
   conducted at WuXi AppTec, was approved under ethical approval ID
   GP01-QD112-2024v1.2 ([293]https://www.wuxiapptec.com). All mouse
   experiments were performed in accordance with Institutional Animal Care
   and Use Committee (IACUC) guidelines and in compliance with the Animal
   Welfare Act, the Guide for the Care and Use of Laboratory Animals and
   the Office of Laboratory Animal Welfare.

   All mice were housed at 23 ± 1 °C on a 12-h light and 12-h dark cycle
   with ad libitum access to food and purified water. Mice were allowed to
   acclimatize on the different diet as described for 7–14 days before the
   experiments. Group assignments were based on their baseline body
   weights, and mice were typically housed in groups, except otherwise
   specified. In the mouse studies, no typical side effects such as
   vomiting, diarrhoea, constipation or physical activity behaviour were
   noted.

   In an 8-week feasibility study, we investigated the potential impact of
   two doses of RT strain on glucose tolerance in mice, all of which were
   8 weeks old at the start of the intervention. The mice that were fed
   with normal chow diet (63.9% kcal% carbohydrate, 20.3% kcal% protein
   and 15.8% kcal% fat) were randomized into groups of 3 or 4 mice. They
   were administered oral gavage twice weekly (3–4 days apart), with
   sterile PBS containing 10% glycerol, live RT2 at doses of 5 × 10^7 (low
   dose) and 5 × 10^8 (high dose) CFU per 100 µl in sterile PBS containing
   10% glycerol and heat-killed RT2 at a dose of 5 × 10^8 CFU per 100 µl
   in sterile PBS containing 10% glycerol, respectively. At week 7 of the
   experiment, an intraperitoneal glucose tolerance test (ipGTT) was
   performed. The mice were fasted for 4 h followed by an intraperitoneal
   injection of glucose (2 g kg^−1 body weight) dissolved in sterile PBS.
   Blood samples from the tail vein were taken at t = 0, 30, 60, 90 and
   120 min for measurement of blood glucose applying glucose strips
   (Bayer, 84167836) and a CONTOUR NEXT EZ meter. All measures were
   performed on the regular laboratory bench to avoid noise and additional
   stress to the mice.

   In the intervention testing the effect of the RT strain at 5 × 10^9 CFU
   per 100 µl in sterile PBS containing 10% glycerol, 8-week-old mice were
   randomized to groups of 8–10 mice. The mice were then fed a high-fat
   diet (HFD, 45% kcal% fat, 35% kcal% carbohydrate and 20% kcal% protein;
   Research Diet D12451i) and provided purified water ad libitum. For
   8 weeks, the animals received oral gavage twice weekly (3–4 days apart)
   with either sterile PBS containing 10% glycerol, live RT2 or
   heat-killed RT2 at 5 × 10^9 CFU per 100 µl in sterile PBS with 10%
   glycerol. At week 7 of the experiment, an ipGTT was performed. Faecal
   samples were collected before gavage and at the end of the experiment,
   then immediately stored at −80 °C until further analysis. Body weights
   were recorded using Analytical Balance Kern ADB 200-4 weight scales.
   Lean and fat mass were determined by quantitative magnetic resonance
   using Echo-MRI 4-in-1 Body Composition Analyser (EchoMRI) at week 6.

   At the end of each mouse study, the animals were fasted for 4 h and
   euthanized with cervical dislocation. Depending on the specific study
   objective, various tissues (liver, brown adipose tissue, inguinal white
   adipose tissue and epididymal white adipose tissue) were immediately
   dissected on ice. The mass of the freshly dissected inguinal or
   epididymal fat tissues was recorded using Analytical Balance Kern ADB
   200-4 weight scales. Inguinal white adipose tissue depots were fixed in
   4% paraformaldehyde and 1× PBS overnight at 4 °C, followed by immersion
   into ethanol (100% w/v) for 24 h before paraffin embedding. The
   remaining tissues were stored at −80 °C for further analysis. For the
   determination of adipocyte size, adipose tissue paraffin sections were
   stained with haematoxylin and eosin (H&E staining; n = 3 in each of the
   three groups). Images were obtained under bright-field microscopy with
   ×100 magnification. A representative image for each group was shown in
   our study. Fat cell size was quantified and compared in the three
   intervention groups using AdipoCount^[294]51, which is an automated
   system for counting adipocytes, using image processing algorithms to
   greatly enhance the efficiency of adipocyte quantification. Femurs were
   precisely dissected and fixed in 4% paraformaldehyde and 1× PBS at 4 °C
   before the microcomputed tomography (micro-CT) analysis. Faecal samples
   from the mice were collected before they were killed. Microbial DNA
   extraction and qPCR assay were performed as previously described using
   primer pairs specifically targeting the RORDEP1 gene (Supplementary
   Table [295]10) for the assessment of the engraftment of RT2 after the
   intervention.

   For mRNA expression analysis by qPCR of thermogenesis, browning and
   inflammation markers, tissue samples were homogenized with a steel bead
   (Qiagen, 69989) and 500 µl of QIAzol Lysis Reagent (Qiagen, 79306) per
   50 mg of frozen tissue. Post-homogenization, the samples were
   centrifuged at 12,000 g for 15 min at 4 °C, and the supernatant was
   collected. RNA was extracted according to the instructions provided by
   the manufacturer of the RNeasy mini kit (Qiagen, 74106), followed by
   measurements of RNA purities and concentrations by NanoDrop 2000/2000c
   spectrophotometers (Thermo Fisher, ND2000CLAPTOP). Reverse
   transcription was done on RNA (1 µg) from last step using the
   High-Capacity RNA-to-cDNA Kit (Thermo Fisher Scientific, 10704217),
   following the provided protocol and heating process. The resulting cDNA
   was then analysed by real-time PCR using the LightCycler 480 System
   (Roche Diagnostics), combined with Precision PLUS Master Mix (Primer
   Design, PPLUS- Roche Diagnostics). For each gene of interest, the PCR
   was performed in white 384-well plates, using the ∆∆Ct method to
   quantify RNA expression levels. Primer sequences for the qPCR are
   listed in Supplementary Table [296]10.

   Proteins were extracted from 50–100 mg of brown adipose tissue using
   RIPA lysis buffer (Millipore, 20-188) with added protease inhibitor
   (Thermo Fisher Scientific, A32953). Protein concentrations were
   quantified using the Pierce BCA Protein Assay kit (Thermo Fisher
   Scientific, [297]A55864). The proteins were then diluted with 4×
   Laemmli protein sample buffer (Bio-Rad, 1610747), heated at 95 °C for
   5 min and separated on a 4–15% Criterion TGX Precast Midi Protein Gel
   (Bio-Rad, 5671084). For immunoblotting, anti-UCP1 (Abcam, ab10983) and
   anti-β-actin antibodies (Abcam, ab115777) were used, the latter serving
   as the housekeeping control. The Gel Doc XR+ System (Bio-Rad) was used
   to visualize the membranes according to the guides from Bio-Rad.

   High-resolution desktop microcomputed tomography imaging (Bruker,
   Skyscan 1172) was applied for micro-CT analysis of mouse distal femurs
   of HFD-fed mice. The parameters for the scan are as follows: X-ray
   voltage of 50 kV, X-ray current of 200 µA, filter of 0.5 mm aluminium,
   image pixel size of 4–5 µm, camera resolution of 1,280 pixel field
   width, tomographic rotation of 180°/360°, rotation step of 0.3–0.5°,
   frame averaging of 1–2 and scan duration of 30–50 min. The cortical
   boundaries (metaphyseal–diaphyseal) were selected with reference to the
   growth plate. A cross-sectional slice was selected as a growth plate
   reference slice, in the following way: moving slice by slice towards
   the growth plate from the metaphysis–diaphysis, a point is reached
   where a clear ‘bridge’ of low-density cartilage (chondrocyte seam)
   becomes established from one corner of the cross-section to another.
   This bridge is established by the disappearance of the last band of
   fine primary spongiosa bone interrupting the chondrocyte seam. This
   landmark allows a reference level to be defined for the growth plate:
   cortical volumes of interest are then defined relative to this
   reference level. The cortical region commenced about 2.15 mm (500 image
   slices) from the growth plate level in the direction of the metaphysis
   and extended from this position for a further 0.43 mm (100 image
   slices). Three-dimensional (3D) morphometric parameters were calculated
   for the cortical selected regions of interest. The 3D parameters were
   based on the analysis of a marching-cubes-type model with a rendered
   surface. Structure thickness in 3D was calculated using the local
   thickness or ‘sphere-fitting’ method, and the structure model index (an
   indicator of the relative prevalence of plates and rods) was derived
   according to the method of Hildebrand and Ruegsegger. The degree of
   anisotropy was calculated by the mean intercept method. Rendered 3D
   models were constructed for 3D viewing of cortical analysed regions.
   Model construction was by the ‘double time cubes’ method, a
   modification of the marching cubes method. The 3D cortical thicknesses
   (Ct.Th, mm) of the femurs collected from the HFD-fed mice were compared
   in this study.

   The intervention testing the effects of the engineered EcN strains
   included 24 HFD-fed DIO mice, aged 20 weeks (initial 8 weeks of age
   plus 12 weeks on the HFD). Each mouse was single housed. From
   acclimatization to the end of the intervention, the mice were fed a HFD
   (60% kcal% fat, 20% kcal% carbohydrate and 20% kcal% protein; Research
   Diets, D12492i) and provided purified water ad libitum. Body weight was
   recorded in the morning of every third day on a Precision Scale KERN
   EMB weight scale. On day −4 of the study (the 13th day of
   acclimatization), regular water was replaced with streptomycin water
   (5 g l^−1). On day 0 of the study, faecal samples were collected from
   each mouse. This marked the beginning of a 4-day oral gavage
   intervention (days 0–3 of the study, 100 µl of the engineered strains
   (5 × 10^10 CFU per 100 µl)) with freshly prepared EcN strains
   administered daily. On day 4, post-gavage, additional faecal samples
   were collected and analysed to confirm stable colonization. To monitor
   the engraftment during and at the end of the study, faeces and
   intestinal contents were collected and dissolved in a pre-weighted
   1.5 ml Dulbecco’s PBS solution (pH 7.0–7.3, Gibco, 14190144) on ice and
   manually homogenized. The weight of faecal sample was recorded. Solids
   were spun down at 500 g and the supernatants (in two technical
   replicates) were serially diluted (10^8 dilution) in the same solvent.
   Dilutions were spotted on lysogeny broth agar supplemented with
   50 μg ml^−1 streptomycin and 50 μg ml^−1 kanamycin. After an overnight
   incubation at 37 °C, engraftment was estimated as CFU ml^−1 per gram of
   faeces. On day 20, the mice underwent an ipGTT. On day 21 of the study,
   the mice were euthanized in a 4–6-h fasted state. Blood, tissue and
   intestinal contents were collected for further analysis.

   Effects on expression thermogenesis marker genes (inguinal white
   adipose tissue) following daily intraperitoneal injections of
   r-RORDEP1, r-RORDEP2 (both at 1 mg kg^−1 body weight) or physiological
   saline for 7 days were explored in mice fed with chow diet at the age
   of 8 weeks (n = 6 mice per group). Tissues were collected on ice at the
   time of death of the mice and stored at −80 °C until analyses. The mRNA
   levels of indicated genes were analysed by qPCR as described in the
   gene expression analysis above.

   To investigate the impact of r-RORDEP1 in regulating blood glucose in
   the diabetic mouse model, male BKS-Leprdb/db/JOrlRj mice (Mus musculus)
   were obtained from Janvier Laboratories at 8 weeks of age, with initial
   body weights ranging from 30 g to 45 g. Upon arrival, the mice were
   singly housed to meet scientific requirements and were maintained on a
   Purina 5008 (FormuLab Diet) regimen. After a 2-week acclimation period,
   the study commenced when the mice reached 10 weeks of age. All animals
   were kept under a controlled 12:12-h light–dark cycle, with lights
   turned off at 3 pm daily. Before treatment, mice underwent
   randomization based on their 4-h fasted blood glucose levels measured
   3 days before the study (day −3). Only mice showing fasting blood
   glucose concentrations exceeding 10 mM were included. These mice were
   then assigned to receive daily intraperitoneal injections of r-RORDEP1
   at a dosage of 2 mg kg^−1 or PBS as a control for a duration of 10 days
   (n = 6 or 7 per group). Throughout the treatment period, body weights
   and blood glucose levels were monitored to assess metabolic changes. On
   day 11, an IPGTT was performed to evaluate glucose tolerance.

   In the comparative study of the effects of r-RORDEP1 and scrambled
   r-RORDEP1 on glucose tolerance, the scrambled r-RORDEP1 sequence was
   generated using GenScript’s peptide screening tools
   ([298]https://www.genscript.com/peptide_screening_tools.html), in which
   permutations of the original RORDEP1 amino acid sequence were performed
   to create multiple scrambled variants. The final scrambled sequence,
   illustrated in Extended Data Fig. [299]7a, was selected based on its
   maintained similarity in identity, secondary structure, hydrophobicity
   and water solubility to the native r-RORDEP1, as shown in Extended Data
   Fig. [300]7b. The scrambled RORDEP1 was recombinantly synthesized using
   the pET30a expression vector with NdeI (CATATG) and HindIII (AAGCTT)
   restriction sites, without biotinylation, using an identical expression
   protocol as used for r-RORDEP1. A total of 45 male C57BL/6J mice at
   8 weeks of age were obtained from Gempharmatech. Upon arrival, the mice
   were singly housed in cages with unique cage numbers for
   identification. The housing environment was maintained under specific
   pathogen-free conditions with controlled temperature (20–24 °C) and
   relative humidity (30–70%), monitored daily. An electronic lighting
   system provided a 12-h light and 12-h dark cycle. Enrichment toys were
   supplied to promote well-being. Mice had ad libitum access to a
   standard chow diet and fresh water. The animals were acclimated in the
   testing facility for 2 weeks before the study. After acclimation, 36
   mice were selected based on body weight and fasting blood glucose
   levels and randomly assigned to 3 groups of 12 mice each. The groups
   included a vehicle control group receiving PBS, a group receiving
   r-RORDEP1 at a dose of 1 mg kg^−1 and a group receiving scrambled
   RORDEP1 at the same dose. On day 1, following a 6-h fasting period,
   each mouse received an oral glucose load of 2 g kg^−1 body weight.
   Immediately thereafter, the mice were administered their respective
   treatments via intraperitoneal injection according to the dosing
   protocol. Tail vein blood glucose levels were measured at 0 (baseline),
   30, 60 and 120 min post-glucose load using a hand-held glucometer
   (Lifescan, Johnson & Johnson Medical Equipment). Measurements were
   performed in duplicate at each timepoint for each mouse to ensure
   accuracy, and the mean value was used for further analysis.

   In the assessment of the plasma half-life of r-RORDEP1, 8-week-old
   chow-diet-fed lean male C57BL/6J mice (n = 4 per group) were
   intraperitoneally injected with either r-RORDEP1 (1 mg kg^−1) or
   sterilized PBS, and blood was collected from the tail vein at indicated
   timepoints. The r-RORDEP1 in plasma was measured by immunoblotting,
   using an anti-His-Tag Antibody (Cell Signal Technology, 2365) to detect
   the 6×His-tag on r-RORDEP1. The pixel intensity of the immunoblotting
   bands of 6×His-tag RORDEP1 and the adjacent background were measured
   using ImageJ^[301]52.

Experiments in rats

   All rat experiments were conducted in accordance with national and
   international guidelines for the care and use of laboratory animals and
   were approved by the relevant regulatory authorities. Studies involving
   intravenous infusion (approval ID: 2023-15-0201-01508) and assessment
   of r-RORDEP1 effects on incretin release (approval ID:
   2023-15-0201-01393) in rats were approved by the Danish Animal
   Experimentation Council. The intraduodenal infusion study was reviewed
   and approved by the Animal Ethics Committee of Chongqing Medical
   University (approval ID: IACUC-CQMU-2024-0036).

   In the intravenous infusion study testing the effect of r-RORDEP1 on
   glucose homeostasis in rats, 12 male lean chow-fed 8-week-old Sprague
   Dawley rats were randomized based on body weight (measured on study day
   −7). In study week −1, intravenous (IV) catheters were implanted in the
   jugular vein (for infusion) and carotid artery (for blood sampling). On
   study day 1, the rats were infused with either vehicle (PBS) or
   r-RORDEP1 (200 pmol kg^−1 min^−1) at a flow rate of 0.25 ml h^−1. Blood
   glucose concentration was measured at t = 0, 15, 30, 60, 90, 120 and
   180 min.

   To assess the impact of r-RORDEP1 on rat incretin release, 48
   8-week-old lean male Sprague Dawley rats were randomly assigned to 4
   groups, with 12 rats per group. Each rat received an oral glucose load
   of 1 g kg^−1, administered over a 2-min period. Immediately following
   glucose administration, rats in the two experimental groups were
   intraperitoneally injected with r-RORDEP1 at a dosage of 0.8 mg kg^−1.
   By contrast, rats in the two control groups received an equivalent
   volume of PBS as a vehicle control. Blood glucose levels were measured
   at baseline, 30 min and 45 min post-glucose administration to evaluate
   the glycaemic response. At designated timepoints of either 30 min or
   45 min after glucose administration, rats were humanely killed. Cardiac
   blood was collected using a syringe and transferred into
   Microvette/Vacuette tubes containing an anticoagulant. To preserve the
   integrity of the analytes, dipeptidyl peptidase-4 and protease
   inhibitors were added to the blood samples immediately upon collection.
   The tubes were then inverted five times to ensure thorough mixing and
   kept at 4 °C until centrifugation at 3,000 g for 10 min. Following
   centrifugation, plasma supernatants were carefully transferred to new
   tubes and promptly frozen on dry ice. The plasma samples were
   subsequently stored at −70 °C until further analysis. Incretin
   hormones, including GIP, GLP1 and PYY, insulin, and glucagon were
   quantified using ELISA-based methods. Specifically, rat plasma GIP
   levels were measured following the manufacturer’s protocol using the
   Rat/Mouse GIP (Total) ELISA kit (EZRMGIP-55K; Merck Millipore), while
   the plasma GLP1 levels were quantified using the MSD V-PLEX Plus GLP1
   Total Kit (catalogue number K1503PG). PYY concentrations were
   determined using a commercially available ELISA kit (Alpco, Mouse/Rat
   PYY ELISA, catalogue number 48-PYYRT-E01.1). Rat insulin levels were
   measured using a homogeneous LOCI/AlphaLISA assay. This assay involved
   the formation of a concentration-dependent bead–analyte–immune complex,
   resulting in luminescent output measured using a PerkinElmer Envision
   reader. The assay used anti-rat insulin monoclonal antibody
   D3E7-conjugated acceptor beads and biotinylated monoclonal antibody
   D6C4, both specific to rat insulin, in conjunction with generic
   streptavidin-coated donor beads. Rat glucagon levels were assessed
   using LOCI/AlphaLISA technology, similarly forming bead–analyte–immune
   complexes that emit luminescence measured by the PerkinElmer Envision
   reader. The assay used in-house-produced anti-glucagon monoclonal
   antibody GLU 1F20-conjugated acceptor beads and biotinylated monoclonal
   antibody GLU 2F7, both targeting glucagon, along with generic
   streptavidin-coated donor beads.

   In the duodenal infusion study, we applied the following protocol: male
   Sprague Dawley rats, 9 weeks old and weighing around 300 g, were
   purchased from the Experimental Animal Center of Chongqing Medical
   University. During the experiments, rats were housed on a 12-h light
   and 12-h dark cycle with ad libitum access to purified water and a
   rodent chow diet (63.9% kcal% carbohydrate, 20.3% kcal% protein and
   15.8% kcal% fat). All rats were acclimatized to the environment for
   7 days before any designated surgical procedures.

   On day 1, duodenal cannulations were performed in rats as previously
   reported^[302]53. In brief, a catheter was inserted into the proximal
   duodenum (approximately 1.5–2 cm downstream of the pyloric sphincter)
   and indwelling catheters were placed into the internal jugular vein for
   duodenum infusion, and intravenous infusion, respectively. Recovery
   from surgery was estimated by monitoring a normalization of food intake
   and body weight stabilization to pre-surgery levels. To ensure similar
   nutritional status, rats were restricted to 20 g of food on the night
   preceding the euglycaemic pancreatic clamp experiments. Rats were
   randomized into six different intervention groups, each consisting of
   six animals. On day 5 after surgery, a euglycaemic pancreatic clamp was
   performed for 4 h. In the euglycaemic pancreatic clamp, endogenous
   secretion of insulin and glucagon is suppressed by intravenous infusion
   of somatostatin while blood glucose is maintained at euglycaemia and
   plasma insulin is kept constant at study basal level by continuous
   infusion of exogenous insulin. An infusion of concentrated glucose is
   given intravenously, the infusion rate being dynamically regulated
   according to frequent monitoring of blood glucose. Thus, continuous
   intravenous infusions of somatostatin (at 3 μg kg^−1 min^−1, Kunming
   Jida Pharmaceutical) and short-acting human insulin (at
   1.2 mU kg^−1 min^−1, Novolin R, Novo Nordisk) via the right internal
   jugular vein were given from t = 0–240 min. r-RORDEP1 was administered
   through duodenal infusion with infusion rates of 0, 60, 90, 120, 150
   and 200 pmol kg^−1 min^−1, respectively at t = 60–240 min of the
   euglycaemic pancreatic clamp. Blood samples from the tail vein were
   taken every 10 min for measurement of blood glucose using a portable
   glucometer. To maintain blood glucose at normoglycaemia, 25% (w/v)
   glucose solution was infused at the dynamic infusion rate of
   t = 0–240 min.

   To explore the impact of r-RORDEP1 on blood glucose control and liver
   transcriptomic and proteomic alterations under physiological
   conditions, we administered r-RORDEP1 via intraduodenal infusion at a
   rate of 200 pmol kg^−1 min^−1 for 3 h. Blood glucose levels were
   measured before and after infusion (t = 0 and 180 min). Following the
   infusion, rats were euthanized, and liver tissues were harvested for
   bulk RNA sequencing analysis, bulk proteomics and phosphoproteomics
   analyses.

   Total RNA was extracted from tissues using Trizol (Invitrogen)
   according to manual instructions. Subsequently, total RNA was qualified
   and quantified using a NanoDrop and Agilent 2100 bioanalyzer (Thermo
   Fisher Scientific). The RNA library construction and subsequent RNA
   sequencing were performed by BGI-Shenzhen, China. First-strand cDNA was
   generated using random hexamer-primed reverse transcription, followed
   by a second-strand cDNA synthesis. Afterwards, A-Tailing Mix and RNA
   Index Adapters were added by incubating to end repair. The cDNA
   fragments obtained from the previous step were amplified by PCR, and
   products were purified by Ampure XP Beads, then dissolved in EB
   solution. The product was validated on the Agilent Technologies 2100
   bioanalyzer for quality control. The double-stranded PCR products from
   the previous step were heated, denatured and circularized by the splint
   oligo sequence to get the final library. The single-strand circle DNA
   (ssCir DNA) was formatted as the final library. The final library was
   amplified with phi29 to make DNA nanoballs, which had more than 300
   copies of one molecule, DNA nanoballs were loaded into the patterned
   nanoarray, and single-end 50-base reads were generated on the BGIseq500
   platform (BGI).

   The sequencing data were filtered with SOAPnuke (v1.5.2), and the clean
   reads were mapped to the reference genome of the
   GCF_000001895.5_Rnor_6.0 genome using HISAT2 (v2.0.4). Bowtie2 (v2.2.5)
   was applied to align the clean reads to the reference coding gene set,
   then expression level of gene was calculated by RSEM (v1.2.12).

   Downstream analysis of RNA sequencing data was executed using an
   in-house-developed script in Rstudio (v4.3.0). In multivariate
   statistical analysis, dimensionality reduction was conducted via
   principal component analysis (PCA) using the ‘prcomp’ function in R
   (v4.3.0). This facilitated the discrimination of sample clusters and
   identification of outliers. Prior to PCA, a variance stabilizing
   transformation was applied to normalize count data distribution across
   samples, ensuring PCA outputs more accurately reflected biological
   rather than extreme values introduced by technical variance. Univariate
   statistical differential analysis was quantified using DESeq2
   (v1.40.2). The ‘lfcShrink’ function of DESeq2 invoked the adaptive
   shrinkage method to mitigate the impact of low counts and high variance
   on log[2](fold change) estimates. All reported fold changes were thus
   shrunken log[2] values, with differentially expressed genes defined by
   an adjusted P ≤ 0.05 and an absolute log[2](fold change) over 1. For
   the data processing and visualization, packaged including dplyr
   (v1.1.4) and tidyr (v1.3.0) and ggplot2 (v3.4.4) were used.

   In the sample preparation for bulk proteomics and phosphoproteomics
   analyses, rat liver tissues were collected following a 3-h
   intraduodenal infusion of r-RORDEP1. Approximately 10–25 mg of liver
   tissue was used for each sample. The tissues were transferred into 2-ml
   centrifuge tubes containing two steel beads. For proteomics samples,
   10 mg of tissue was homogenized in an appropriate volume of SDS L3
   lysis buffer supplemented with a 1× protease inhibitor cocktail
   containing EDTA. For phosphoproteomics samples, 25 mg of tissue was
   homogenized in 300 μl of Milli-Q water. Homogenization was performed
   using a grinder.

   For proteomics samples, the homogenate was centrifuged at 25,000 g for
   15 min at 4 °C to collect the supernatant. DTT was added to the
   supernatant to a final concentration of 10 mM, and the samples were
   incubated in a water bath at 37 °C for 30 min. For phosphoproteomics
   samples, 50 μl of 10% SDS was added to the homogenate, followed by DTT,
   to a final concentration of 10 mM. The samples were incubated at 95 °C
   in a metal bath for 10 min.

   Subsequent steps were identical for both sample types. Iodoacetamide
   was added to each sample to a final concentration of 55 mM, and the
   mixtures were incubated in the dark for 45 min. Proteins were
   precipitated by adding cold acetone at a ratio of 1:5 (protein
   solution) and incubating at –20 °C for 2 h. The precipitated proteins
   were collected by centrifugation at 25,000 g for 15 min at 4 °C, and
   the supernatant was discarded. The protein pellets were air-dried and
   resuspended in an appropriate volume of SDS-free lysis buffer L3. The
   samples were homogenized again using a grinder to promote protein
   solubilization. Finally, the samples were centrifuged at 25,000 g for
   15 min at 4 °C, and the supernatants containing the protein solutions
   were collected. Protein concentrations were measured using the Bradford
   assay.

   For each sample, 100 μg of protein solution was diluted fourfold with
   50 mM ammonium bicarbonate (NH₄HCO₃). Trypsin was added at a
   protein-to-enzyme ratio of 40:1 (2.5 μg of trypsin per 100 μg of
   protein), and the mixtures were incubated for digestion at 37 °C for
   4 h. The resulting peptides were desalted using Strata X columns and
   vacuum dried.

   For library generation, a high-pH reverse-phase fractionation was
   performed. Equal amounts of peptides from all samples were pooled and
   diluted with mobile phase A (5% ACN, pH 9.8). The pooled sample was
   subjected to high-pH reverse-phase chromatography using a Shimadzu
   LC-20AB HPLC system coupled with a Gemini C18 column (5 μm,
   4.6 × 250 mm). Peptide separation was performed at a flow rate of
   1 ml min^−1 using the following gradient: 5% mobile phase B (95% ACN,
   pH 9.8) for 10 min; 5% to 35% B over 40 min; 35% to 95% B over 1 min;
   maintained at 95% B for 3 min; and equilibrated at 5% B for 10 min. The
   elution was monitored at 214 nm, and fractions were collected every
   minute. Ten fractions were combined and freeze-dried.

   The LC–MS/MS analysis involved two main steps: data-dependent
   acquisition (DDA) for spectral library construction and DIA for peptide
   quantification. The nano-LC protocol used for both DDA and DIA modes
   was identical. Dried peptide samples were reconstituted in mobile phase
   A (2% acetonitrile, 0.1% FA) and centrifuged. The supernatants were
   injected into an ultra-high-performance liquid chromatography system
   for separation. Peptides were first trapped and desalted on a
   pre-column, then separated on an analytical column at a flow rate of
   500 nL/min using a specified gradient.

   For DDA library construction in bulk proteomics analysis, peptides were
   ionized using a nano-electrospray ionization source and analysed on an
   Orbitrap Exploris 480 mass spectrometer operated in DDA mode. Mass
   spectrometry parameters were optimized to enhance peptide
   identification. For DIA quantification, the mass spectrometer was
   operated in DIA mode to acquire comprehensive peptide ion data across
   the mass range. Instrument settings were adjusted to achieve
   high-resolution MS/MS scans suitable for accurate quantification.

   In the phosphoproteomics analysis, the separated peptides were analysed
   using a timsTOF Pro mass spectrometer (Bruker Daltonics) operated in
   DDA mode for spectral library construction. This instrument combines
   trapped ion mobility spectrometry with time-of-flight mass
   spectrometry, providing enhanced separation and identification of
   phosphopeptides due to its high-resolution and ion mobility
   capabilities. Specific acquisition parameters were optimized to
   maximize the detection and characterization of phosphorylated peptides.

   DDA data were processed using the Andromeda search engine within
   MaxQuant software to identify peptides and proteins, constructing
   spectral libraries for both bulk proteomics and phosphoproteomics
   analyses. DIA data were analysed using retention time calibration with
   indexed retention time peptides. A target–decoy approach was used to
   control the false discovery rate at 1%, ensuring reliable
   identification and quantification.

   For phosphoproteomics data, phosphorylation sites were filtered based
   on the localization probability reported by MaxQuant, retaining only
   those sites with a localization probability of at least 0.75 to ensure
   confidence in site-specific phosphorylation analysis.

   Differential expression analysis was conducted using the MSstats
   package in R, which applies a linear mixed-effects model to evaluate
   significant differences in protein or phosphoprotein abundance between
   samples. Data were preprocessed according to predefined comparison
   groups, and significance testing was performed based on the model.
   Proteins and phosphoproteins with an adjusted P value of less than 0.05
   were considered significantly differentially expressed.

   Lastly, to further evaluate the effect of r-RORDEP1 on glucose
   metabolism in vivo, titrated glucose was used as a tracer to observe
   alterations in hepatic glucose production and peripheral glucose
   disposal in a separate experiment under euglycaemic pancreatic clamp.
   [3-H^3]-glucose was administered with a primary intravenous bolus of
   6 μCi at time 0 min followed by a continuous intravenous infusion of
   0.2 μCi min^−1 throughout the experiment of 240 min via the right
   internal jugular vein. Recombinant RORDEP1 at a rate of
   200 pmol kg^−1 min^−1 or PBS was infused into the duodenum from 0 to
   240 min. Blood sampling was performed from the carotid artery at
   t = 60–90 and t = 220–240 min. These samples were subsequently
   centrifuged to separate plasma and stored at −80 °C for the analysis of
   [3-H^3]-glucose specific activity. Hepatic glucose production rate and
   whole-body glucose uptake rate were calculated using specific activity
   of the [3-H^3]-glucose and the recordings of the dynamic 25% glucose
   infusion rates^[303]54.

   The tracer dilution principle is central to the isotope calculations.
   When introducing a known quantity of a labelled tracer (in case,
   [3-H^3]-glucose) into the bloodstream, it mixes with the body’s
   existing glucose. As the body produces glucose (endogenous glucose
   production) plus exogenous infusion of glucose, these glucose
   compartments dilute the concentration of the labelled tracer in the
   blood. In the steady phase of a euglycaemic pancreatic clamp, the rate
   of this dilution reflects the rate of endogenous glucose production
   plus exogenous infusion of glucose. The equal removal of labelled and
   unlabelled glucose from the bloodstream by tissue uptake minimally
   affected the dilution of the labelled tracer.

   Thus, the calculation of hepatic gluconeogenesis was based on three
   formulas: formula 1, calculation of the glucose appearance rate as the
   sum of endogenous glucose production and glucose infusion rate; formula
   2, equation of the glucose appearance rate to the glucose disappearance
   rate, equal to the glucose uptake rate; and formula 3, estimation of
   the tracer infusion rate ÷ glucose appearance rate = specific activity
   of the tracer in blood. In formula 3, the tracer infusion rate is the
   known component at which we infused the isotope-labelled glucose into
   the bloodstream. Specific activity is defined as the ratio of the
   labelled glucose ([3-H^3]-glucose) to the total amount of the glucose
   (both labelled and unlabelled glucose) in the blood.

   Finally, the following two formulas were used in the calculation of the
   whole-body glucose uptake rate and hepatic glucose production: (1) the
   whole-body glucose uptake rate was calculated as tracer infusion
   rate ÷ specific activity of [3-H^3]-glucose in blood and (2) the HGP
   rate was calculated as whole-body glucose uptake rate ÷ glucose
   infusion rate.

Statistical analysis

   All the presented data were obtained from biologically independent
   samples. Graphing and statistical analysis were performed using
   GraphPad Prism (version 10.1.0). Area under the curve (AUC) was
   calculated using the composite trapezoidal method as described
   previously^[304]55. Full descriptions of statistical tests were
   provided in the figure legends. For two-group comparisons, two-tailed
   Student’s t-test was applied. For multiple group comparisons, one-way
   ANOVA was performed. For comparisons at each timepoint, two-way ANOVA
   was used, followed by correction for multiple testing. Data were
   presented as mean ± s.e.m. unless otherwise stated. P values less than
   0.05 were considered statistically significant.

Reporting summary

   Further information on research design is available in the [305]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [306]Supplementary Information^ (7.2MB, pdf)

   Supplementary Figs. 1–16 and References.