Experiments were undertaken under governmental project license (Health Products Regulatory Authority – AE18982/P084). Fourteen-week-old adult male ZDSD rats (n=35) and Sprague Dawley (SD, n=6) control rats (Crown BioScience) were provided with water and Purina 5008 rodent chow (Nestle Purina, St. Louis, MO). Body weight and glycaemia-matched ZDSD rats were allocated to either a sham-operated disease control (SHAM, n = 9), or one of two treatment groups: RYGB surgery (RYGB) and RYGB surgery plus fenofibrate, metformin, ramipril, and rosuvastatin (RYGB-FMRR). In total, 26 ZDSD rats underwent RYGB surgery, with 8 dying in the intra- or post-operative periods. RYGB was thus associated with a mortality rate of 30.8%. Causes of death included primary intraoperative gastrointestinal haemorrhage (n=1) and anastomotic complications (n=5, including haemorrhage and leakage). No abdominal pathology was identified on post-mortem of two rats who died after RYGB. Of the 18 remaining RYGB-operated ZDSD rats, n=9 each were assigned to the RYGB alone and RYGB-FMRR groups. No mortalities occurred in the SD or SHAM groups.

Rats treated with RYGB-FMRR received 100 mg/kg fenofibrate (Mylan Pharma, Canonsburg, PA), 300 mg/kg metformin (Teva Pharma, Petah Tikva, Israel), 1 mg/kg ramipril (Sanofi, Paris, France), and 10 mg/kg rosuvastatin (Teva Pharma). After RYGB, rats were transitioned from a liquid to a semi-solid and then to a regular chow diet over a two-week period to allow for anastomotic healing. As the above medications were incorporated into daily chow rations, they were introduced two weeks after RYGB once rats were established on a regular chow diet. Metformin monotherapy was introduced for the first two days to monitor for adverse responses, including anorexia. The remaining medications (fenofibrate, ramipril, and rosuvastatin) were commenced thereafter when no adverse response was observed. Medications were administered at doses which have been shown to be renoprotective in monotherapy in rat models of hypertensive or diabetic renal injury (32–35).

Body weights and mid-morning plasma glucose levels (Freestyle Optium Neo, Abbott Laboratories, Chicago, IL) were examined on a weekly basis before and after intervention. Animals were euthanized and tissue (renal cortex, liver, and epididymal fat) collected after an 8-week post-intervention period. The study design, including timing of interventions and sample collection, is summarised in Figure 1 .

Sham surgeries were performed in ZDSD animals at 25 weeks of age ( Figure 1 ). RYGB surgeries were performed at 29 weeks of age in the RYGB-FMRR group and at 30 weeks of age in the RYGB group ( Figure 1 ). In light of the animal husbandry requirements in the postoperative setting, as well as the fact that all RYGB surgeries were performed by a single surgeon (M.A.), it was not possible to perform all RYGB surgeries in a single week and the procedures were thus staggered. One week prior to surgery, glycaemic control was optimized with daily subcutaneous injection of insulin degludec (Tresiba^®, Novo Nordisk) to achieve a fasting plasma glucose below 12 mmol/L. Animals were anaesthetized with isoflurane and administered a pre-operative prophylactic antibiotic, enrofloxacin 5mg/kg s.c. (Baytril, Bayer). In the SHAM group, a midline laparotomy was performed followed by closure. In RYGB groups, the jejunum was transected 15 cm distal to the duodenum and the proximal stump side-to-side anastomosed to the distal jejunum 25 cm from the ileocaecal valve. The stomach was transected 3 mm from the gastro-esophageal junction, creating a small gastric pouch, which was then end-to-side anastomosed to the distal stump of jejunum. The remnant stomach was closed, forming standardized biliopancreatic and alimentary limbs and a common channel. Buprenorphine (Animalcare Limited) analgesia was provided at 0.01-0.05 mg/kg s.c. every 6 hours for the first 2 post-operative days and as required thereafter.

As outlined in Figure 1 , urine samples were collected over a duration of 16 hours at baseline (26 weeks of age) and at 4-week follow-up (33 weeks of age in the RYGB-FMRR group and 34 weeks of age in the RYGB group). Urinary concentrations of albumin were examined by ELISA (K15162C Meso Scale Discovery, Rockville, MD). Serum samples were collected at 8-week follow-up. Serum cholesterol and triglycerides were measured using an Atellica^® Solution Immunoassay and Clinical Chemistry Analyser (Siemens Healthineers).

Ten-micron thick sections of formalin-fixed paraffin-embedded kidney were used for haematoxylin and eosin staining, while five-micron thick sections were used for immunohistochemical staining. Scanned haematoxylin and eosin-stained sections were used to measure glomerular area in QuPath (36), from which glomerular volume was calculated using the Weibel and Gomez formula (37). Thirty glomerular tufts per sample were manually annotated at random throughout the renal cortex, from a minimum of six animals per experimental group. Kidney sections were stained with anti-ACOX1 antibody (ab184032, Abcam) using the VECTASTAIN Elite ABC-HRP peroxidase kit (PK-6200, Vector Laboratories). An IgG isotype control (ADI-950-231-0025, Enzo Life Sciences) and no antibody control were used to confirm the specificity of staining. All slides were digitized using an Aperio AT2 Digital Slide Scanner (Leica Biosystems).

Glutaraldehyde-fixed and 1% osmium tetroxide post-fixed renal tissue was dehydrated and infiltrated with EPON™ Epoxy Resin. Ultra-thin sections were prepared and examined by transmission electron microscopy (Tecnai™ G2 12 BioTWIN). Images were analyzed using ImageJ software (NIH, https://imagej.nih.gov/ij/). Data were acquired from both glomeruli and proximal tubules of six rats from each of the four experimental groups.

All glomerular ultrastructural measurements were recorded from six separate capillary loops, representative of at least three separate glomeruli per sample. Podocyte foot process frequency (PFPF) was measured at 9900x by determining the number of podocyte foot processes (FPs) per unit length (8μm) of glomerular basement membrane (GBM). Six measurements were recorded per sample. GBM thickness was measured according to the Haas method at 20500x (38). Podocyte foot process diameter (PFPD) was measured as a reciprocal of PFPF at 20500x. Twenty-four GBM thickness and PFPD measurements were recorded per specimen.

Mitochondrial roundness was assessed as a marker of mitochondrial stress and measured in thirty images of proximal tubular cells per animal, fifteen each from the pars convoluta and pars recta segments. The pars convoluta and pars recta sections of the proximal tubule were distinguished by their ultrastructural morphometry and mitochondria contained therein imaged separately (39, 40). For each of the pars convoluta and pars recta sections, fifteen non-overlapping images from three distinct regions (five images/region) were acquired at 16500x per sample. Each image contained a minimum of five mitochondria for quantification. Images were captured in the basal region of pars convoluta cells for consistency, adjacent to the tubular basement membrane. Longer mitochondria which ran off the edge of the image, and for which less than 2μm of their course was captured by the image window, were excluded from analysis. Mitochondrial parameters (area, perimeter, major and minor axis, roundness, and aspect ratio) were measured using the freehand line selection tool of ImageJ. Roundness is the inverse of the aspect ratio (major axis/minor axis length) and is calculated as: 4*area/(π*major axis²).

RNA was extracted from renal cortex, liver, and epididymal fat pads using an RNeasy Plus Mini Kit (Qiagen). The concentration and purity of RNA samples were determined using a Nanodrop™ 2000 Spectrophotometer (Thermo Fisher Scientific) and RNA sample integrity assessed using Agilent RNA 6000 Nano kits (Agilent Technologies).

For renal cortical RNA sequencing, RNA library preparation was carried out using the TruSeq Stranded Total RNA Library Prep Gold^® kit (Cat. No. 20020598, Illumina). Libraries were sequenced on the Illumina NovaSeq 6000^® platform in a paired-end fashion at a read length of 2x100bp. Sequencing fastq files and raw counts of aligned reads have been deposited in GEO (accession number GSE147706).

For quantitative reverse-transcription polymerase chain reaction (qRT-PCR) analyses, samples were treated with DNase I and complementary DNA (cDNA) synthesized using SuperScript™ II Reverse Transcriptase Kit (Invitrogen). mRNA expression of Pdk4 in renal cortex and of Acox1, Ehhadh, and Acaa2 in three tissue depots (renal cortex, liver, and epididymal fat) was quantified using beta-actin as the endogenous reference gene (TaqMan^® Gene Expression Assays, Thermo Fisher Scientific and QuantStudio 7 Flex System, Applied Biosystems). Comparative analysis was performed using the ΔΔCt method (41), with SD animals serving as calibrators.

The quality of the raw renal cortical RNA sequencing fastq files was analyzed using FastQC (0.11.2) (42). Quality filtering of reads and adapter removal was performed using Trim Galore (0.4.0) together with Cutadapt (1.9) (43, 44). The data was mapped with STAR (2.5.2b) towards the rat reference genome, rn5 (45). Read quantification was performed with featureCounts (1.6.4) (46).

Further analysis was performed using the R statistical programming language (4.0.5) (47). Differential expression analysis was performed using DESeq2 (48). The data was normalized by size factors. A negative binomial generalized linear model was fitted to the normalized data, with the Wald statistic used to identify differentially expressed genes. The p-values were adjusted for multiple testing with the Benjamini-Hochberg procedure. A regularized log (rlog) transformation was subsequently applied to gene expression counts.

Clustering by principal component analysis was performed using rlog gene expression counts and plotted by experimental group using the R package factoextra (49). Volcano plots were created to highlight the most strongly changed transcripts between groups. Differentially expressed transcripts between experimental groups, considered as those with an absolute fold-change ≥1.3 and adjusted p-value <0.05, were used as the input for functional enrichment analyses. Pathways (Reactome database) and gene ontology terms over-represented between all experimental groups were examined using the function ‘compareCluster’ in the R package clusterProfiler (50). Pathway over-representation analysis between the RYGB-FMRR and RYGB groups was performed using the function ‘enrichPathway’ in the R package ReactomePA (51). Upstream regulator analysis was performed using Ingenuity Pathway Analysis (IPA) (52).

After converting rat ENSEMBL gene identifiers to human ENTREZ gene identifiers using the R package biomaRt (53), the abundance of renal cortical immune and stromal cell populations was estimated by inputting rlog gene expression counts to the function ‘MCPcounter.estimate’ from the Microenvironment Cell Populations-counter (MCP-counter) R package (54). Cell abundance estimates were subsequently plotted on a heatmap using the R package pheatmap (55).

To interrogate localization of transcripts differentially expressed between the RYGB and RYGB-FMRR groups, we accessed a publicly available human kidney single-nucleus RNA-sequencing dataset containing samples from 3 individuals with early DKD (56). Raw gene expression counts and cell assignments were obtained. The data were analyzed according to standard single cell clustering workflows in the R package Seurat, including normalization, identification of variable features between cells, scaling, dimensionality reduction, cell clustering, and ultimately the assignment of identity to clusters of cell types (57). Assignment of cluster identity was cross-checked with the human diabetic kidney dataset on the Kidney Interactive Transcriptomics website (http://humphreyslab.com/SingleCell/), which is the same dataset analyzed herein. Average expression of each transcript across the defined cell types was calculated using the function ‘AverageExpression’. Transcripts differentially expressed between the RYGB and RYGB-FMRR groups, after conversion to their human orthologs (53), were intersected with the human diabetic kidney gene expression matrix and plotted on a heatmap to explore their cell-specific expression patterns (55). Violin plots of cell type expression of three transcripts, Acox1, Ehhadh, and Acaa2, were generated with the Seurat function ‘VlnPlot’.

Further interrogation of the localisation of transcripts differentially expressed between the RYGB-FMRR and RYGB groups was performed using a proteomics dataset of 14 rat tubular epithelial cell types (58), which offered improved granularity for predicted localisation of transcripts along the renal tubule. The proteomics data was downloaded from the Kidney Tubules Expression Atlas website (https://esbl.nhlbi.nih.gov/KTEA/) and imported into RStudio (47). Transcripts differentially expressed between the RYGB-FMRR and RYGB groups were intersected with the rat tubular epithelial cell protein expression matrix and plotted on a heatmap (55). Line plots of cell type expression of three proteins, ACOX1, EHHADH, and ACAA2, were generated.

As the four medications were provided concurrently to rats in the RYGB-FMRR group, we employed a network pharmacology approach to discern contributions of individual medications to transcriptomic differences between the RYGB-FMRR and RYGB groups. We also explored contributions of individual PPAR isotypes in this regard, given the over-representation of PPAR-governed mitochondrial and peroxisomal FAO pathways observed between the two groups.

Genes responsive to FMRR medications (fenofibrate, metformin, ramipril, and rosuvastatin) and PPAR isotypes (alpha, beta/delta, and gamma) were obtained using IPA (52). Separate lists of the four medications, alongside their corresponding cardinal drug targets, as well as the three PPAR isotypes were generated in IPA. For each list, a network was grown using the ‘Grow’ tool in the ‘Build’ section of the ‘My Pathways’ interface. An additional network was grown for all four medications and their cardinal drug targets simultaneously to create a network visualisation of medication-responsive genes contained within the RYGB-FMRR vs RYGB differentially expressed gene (DEG) list. Molecules within the network were limited to RYGB-FMRR vs RYGB DEGs when growing the network in IPA. Only experimentally observed relationships were permitted. Drug, chemical, disease, and function categories were excluded.

The data underlying the individual networks constructed for each medication and PPAR isotype were exported as a.txt file and imported into RStudio (47). Orthologous rat genes were obtained by converting the rat Entrez ID (contained within IPA export) to rat gene symbol using the R packages AnnotationDbi and org.Rn.eg.db (59, 60). The target gene data for each medication and PPAR isotype was intersected with the RYGB-FMRR vs RYGB DEG list on the basis of gene symbol.

Doughnut plots were generated to illustrate the number and percentage of medication- and PPAR-responsive genes identified in IPA, stratified by medication type/PPAR isotype (inner layer) as well as presence in or absence from the RYGB-FMRR vs RYGB DEG list (outer layer). For the subsets of medication- and PPAR-responsive genes present in the RYGB-FMRR vs RYGB DEG list, Venn diagrams were created using the R package ggvenn to illustrate the overlap and separation in transcripts responsive to each of the four medications and each of the three PPAR isotypes (61).

To interrogate cellular localisation along the renal tubule, medication- and PPAR-responsive transcripts contained within the RYGB-FMRR vs RYGB DEGs were intersected with a rat tubular epithelial cell protein expression matrix (58). Protein abundance across all non-proximal tubular cell types was collapsed into a ‘Rest of tubule’ category by obtaining row-wise means across the relevant cell types. Cell type-specific abundance of medication- and PPAR-responsive transcripts present in the RYGB-FMRR vs RYGB DEGs was plotted on heatmaps (55). Transcripts clustered on the heatmaps based on relative abundance along the renal tubule; this information was extracted from the heatmap dendrograms and transcripts were accordingly classified as belonging to one of two categories, the proximal tubule or the rest of the renal tubule, based on site of maximal abundance. Doughnut plots were subsequently generated for the medication- and PPAR-responsive transcripts present in the RYGB-FMRR vs RYGB DEGs to summarise the number and percentage of transcripts stratified by fenofibrate- or PPARα-responsiveness (inner layers), presence or absence from the tubular epithelial cell proteomics dataset (middle layers), and localisation in either the proximal tubule or the rest of the renal tubule (outer layers).

Pathway over-representation analysis was performed for the subsets of fenofibrate- and PPARα-responsive transcripts present in the RYGB-FMRR vs RYGB DEGs using the function ‘enrichPathway’ in the R package ReactomePA (51), with results presented on a dotplot.

¹H-NMR spectroscopy was performed on timed urine samples obtained at baseline and at 4 weeks after intervention according to standard Bruker In Vitro Diagnostics for research (IVDr) methods. Urine samples were thawed at room temperature for 20 min before a brief spin at 2000 g at 4°C for 10 min. NMR samples for 5 mm SampleJet racks were prepared by mixing 9 parts urine with 1 part urine buffer (1.5 M KH₂PO₄ pD 6.95, 0.5% w/v NaN₃, 0.1% w/v 3-trimethylsilyl propionic-2,2,3,3 acid sodium salt D4 (TSP-d4) in 99.8% D₂O) using a SamplePro Tube L liquid handling robot (Bruker BioSpin). The temperature was kept at 279 K throughout the sample preparation process. ¹H-¹H total correlation spectroscopy (TOCSY) POM balls were added to tube caps in the finished sample tube rack before placing the rack in the cooled SampleJet sample changer on the spectrometer. 1D Nuclear Overhauser Effect Spectroscopy (NOESY), 1D Carr-Purcell-Meiboom-Gill (CPMG) and 2D J-resolved experiments were acquired for each sample with a 600 MHz Bruker Avance III HD spectrometer at 300 K equipped with a 5 mm BBI room temperature probe, using the pulse sequences ‘noesygppr1d’, ‘cpmgpr1d’ and ‘jresgpprqf’, respectively, according to the IVDr SOP. Urine samples were randomised during sample preparation such that samples from each experimental group were evenly distributed during data acquisition. Pooled samples containing aliquots of samples from each of the study groups were included as an internal quality control. TSP-d4 was used for internal chemical shift referencing.

To facilitate metabolite annotation, a set of 2D experiments on four selected samples were acquired on an Oxford 800 MHz magnet equipped with a Bruker Avance III HD console, a 3 mm TCI cryoprobe and a cooled SampleJet sample changer. ¹³C-HSQCs were acquired using the pulse sequence ‘hsqcedetgpsisp2.3’ using spectral widths of 20 and 90 ppm in the direct and indirect dimensions, respectively, collecting 64 scans per increment for a total of 512 increments and 2048 data points. The acquisition time was 63.9 and 14 ms for the direct and indirect dimensions, respectively, and the relaxation delay was 1.5 s. ¹H-¹H-TOCSYs were acquired using the pulse sequence ‘dipsi2esgpph’ with sweepwidths in both dimensions of 12 ppm, collecting 16 scans per increment into 512 increments and 8192 data points. Acquisition times were 0.426 s and 26.6 ms for the direct and indirect dimensions, respectively. The TOCSY transfer delay was 60 ms and the relaxation delay between scans was 1 s. ¹H-¹H-COSYs were acquired with the pulse sequence ‘cosygpppqfpr’. Sweepwidths were 13.95 ppm in both dimensions; 4 scans per increment were collected to a total of 1024 increments and 2048 data points. The acquisition time was 92 ms and the relaxation delay 2 s. All 2D spectra were referenced to TSP-d4.

The 1D NOESY spectra destined for peak picking and multivariate analysis were zero-filled twice before Fourier transformation into 132k data points, including addition of 0.3 Hz exponential line-broadening and referencing to TSP-d4. Spectra were processed in TopSpin3.5pl7 (Bruker BioSpin). The 1D NOESY spectra were loaded into Matlab using RBNMR (62). Baseline correction of the spectra was performed due to the high urinary glucose concentrations present in untreated ZDSD rats using the command ‘msbackadj’ with window size set to 1000, quantile set to 0.1, and stepsize set to 500 (63). Further analysis of NMR data was performed using the R statistical programming language (4.0.5) (47). Spectra and parts per million (ppm) chemical shift values were imported and processed to a peak intensity matrix according to a standard workflow using the R package Speaq (64). Peak detection was performed using a Mexican hat wavelet method implemented by the function getWaveletPeaks. Detected peaks were aligned and grouped to a single ppm index value using the function PeakGrouper. Illustrative examples of raw spectra, peak detection, and peak grouping/alignment using Speaq for two peaks of interest are presented in Supplemental Figure 1 . Silhouette values were calculated as a metric of the quality of peak grouping using the function SilhouetR. Peak groupings with a silhouette value less than 0.6 were removed and the peaks regrouped with the function regroupR – this process was repeated iteratively until all peak groupings had a silhouette value ≥0.6. Peak filling was performed to detect peaks that may have been missed during the first round of peak detection. Finally, a peak intensity matrix was built with grouped peaks (identified by their ppm shift values) as columns and samples as rows. A probabilistic quotient normalisation (PQN) was applied to the peak intensity matrix (65, 66), which was subsequently used as the input for multivariate statistics. Annotation of processed spectra was performed using Chenomx 8.6 software (Chenomx Inc.), the Human Metabolome Database (HMDB), and the Biological Magnetic Resonance Data Bank (BMRB) (67–69).

Clustering by principal component analysis was performed using PQN-normalised NMR peak intensities and plotted by phenotype as biplots along principal components 1-4 using the R package factoextra (49). To elucidate differences in the urinary metabolome between RYGB and RYGB-FMRR rats at 4 weeks after intervention, the R package MUVR was used to fit a multivariate random forest (RF) classification model to a PQN-normalised NMR peak intensity matrix for post-intervention samples from RYGB and RYGB-FMRR rats (70). A response vector indicating experimental group assignment was inputted to the supervised model. The MUVR algorithm minimises overfitting in multivariate modelling by performing recursive elimination of the least informative variables in a repeated double cross-validation procedure (70). The following modelling parameters were used as recommended: nOuter=5 (number of outer cross-validation segments, to ensure both classes were present in all model segments), nRep=100 (number of model repetitions), and varRatio=0.85 (proportion of variables maintained in the data per model iteration during variable elimination) (70).

MUVR returns three consensus models (min, mid, and max) with similar fitness (70). The max RF model, which attempts to consider all relevant predictors without compromising classification performance, was selected to identify as many urinary NMR peaks relevant to classifying RYGB and RYGB-FMRR status as possible. Model stability across 100 repetitions was ensured by inspecting the number and proportion of selected variables as well as the number of classifications, per repetition and cumulatively; model convergence occurred by 20 model repetitions. The number of model misclassifications was used to assess model performance. Additional performance metrics (area under the curve, sensitivity, and specificity) were calculated using the R package caret (71). Mean decrease in Gini index was used to rank variable (urinary ¹H-NMR peak) importance to RF model classification.

Multi-omic integration of the renal cortical transcriptome and urinary metabolome was performed using the DIABLO (data integration analysis for biomarker discovery using latent variable approaches for omics studies) framework in the R package mixOmics (72, 73). A supervised, sparse partial least squares-discriminant analysis (PLS-DA) model was fit to rlog gene expression counts and annotated, PQN-normalized urinary ¹H-NMR peaks from 4 weeks after intervention. Lowly expressed transcripts were removed from the gene expression count matrix derived from RNA-Seq to reduce the number of inputted transcripts to ~10,000 as recommended to reduce computational time (72). This was performed by removing transcripts for which all samples had an rlog count value <7.

Several aspects of the sparse DIABLO model were tuned to identify a gene-metabolite signature distinguishing the RYGB-FMRR group from the other experimental groups, including the number of model components, the number of features to select from each omics dataset for each model component, and the design matrix (ranging between 0-1 and specifying the extent to which datasets should be connected to maximise the covariance beween components) (73). The following modelling parameters were used after tuning: ncomp=3 (number of model components); number of transcripts to consider for each of the 3 model components: 50; number of metabolites to consider for each of the 3 model components: 10, 20, and 10; and a design matrix value of 0.3 to maximise discrimination between RYGB-FMRR rats and the other experimental groups. Plots of variable loadings (importance) along the model components were generated. A network visualisation of the gene-metabolite signature distinguishing RYGB-FMRR rats from the other experimental groups was generated in mixOmics and edited in Cytoscape (3.7.2) after export using the R package RCy3 (72, 74, 75).

The relationships between changes in kidney structure with changes in renal cortical transcript expression and urinary metabolite abundance were investigated. Pearson correlation matrices between mean values for histological and ultrastructural parameters and rlog gene expression counts as well as PQN-normalised urinary NMR peaks from 4 weeks post-intervention were constructed on a per animal basis. Gene-structure correlations for transcripts which belonged to enriched pathways between RYGB and RYGB-FMRR rats by over-representation analysis were extracted. Metabolite-structure correlations for selected metabolites which were differentially abundant between RYGB and RYGB-FMRR rats were extracted. Correlation matrices were plotted using ggcorrplot (76).

Study endpoints, statistical tests by which they were analyzed, and location within the manuscript are presented in Table 1 . Percentage delta change in metabolic parameters (body weight and plasma glucose) and urinary albumin excretion rates (natural logarithm-transformed) were calculated. Statistical analyses were performed using the R package rstatix in RStudio (R version 4.0.5) (47, 77). For tests involving comparisons between more than two groups, a Benjamini-Hochberg multiplicity correction of p–values was applied. P <0.05 was considered statistically significant. Study endpoints are plotted as violin plots or boxplots, with individual data points for each animal superimposed (78). For histological and ultrastructural data, individual measurements for each animal within each group are plotted to provide additional insight into data distribution (78).

Body weight decreased from 570.8±23.0 to 452.7±32.2 g (mean±SD) following RYGB (p<0.001) and from 564.0±20.4 to 410.0±24.2 g following RYGB-FMRR (p<0.001) ( Table 2 ). Weight loss was greater in the RYGB-FMRR group compared with the RYGB group (-27.3±3.8 vs -20.7±4.7%, p=0.01).

Comparing pre- and post-intervention plasma glucose concentrations, the SHAM group deteriorated by 84.3 ± 95.9% (13.3±6.4 vs 23.2±11.5 mmol/L, p=0.06). Over the same timeframe, RYGB-operated animals improved by 15.7±25.5% (13.0±4.3 vs 11.7±7.5 mmol/L, p=0.30) and the RYGB-FMRR group improved by 50.5±13.0% (13.4±4.9 vs 6.2±1.7 mmol/L, p=0.007). Improvements in glycaemia were greater in the RYGB-FMRR group compared with RYGB-operated animals (p=0.02). Serum cholesterol and triglycerides were elevated in SHAM compared with SD rats. Compared with SHAM rats, RYGB and RYGB-FMRR lowered both serum cholesterol and serum triglycerides, with no differences observed between both interventions (p=0.33 for cholesterol; p=0.81 for triglycerides).

Comparing pre- and post-intervention log UAER, the SHAM group deteriorated by 37.2±30.3% (4.8±0.8 vs 6.8±2.2 μg/hour, p=0.03). RYGB-operated animals improved by 10.6±22.8% (5.1±0.5 vs 4.5±1.2 μg/hour, p=0.32) and the RYGB-FMRR group improved by 36.4±17.5% (5.0±0.8 vs 3.1±0.9 μg/hour, p=0.003). Improvements in albuminuria were greater in the RYGB-FMRR group compared with RYGB-operated animals (p=0.03).

Principal component analysis of kidney RNA-Seq data identified discrete shifts across groups ( Figure 2A ). RYGB-FMRR altered more transcripts (n=1982 vs n=987) and corrected more disease-associated transcripts (i.e., those that were altered between SD and SHAM, n=871 vs 453) than RYGB. Lists of differentially expressed genes between the study groups are available at: https://osf.io/cf7v5/. Volcano plots emphasize that RYGB-FMRR induced more transcripts than RYGB ( Supplemental Figure 2 ).

Both RYGB and RYGB-FMRR downregulated cell cycle and fibrosis pathways while restoring biological oxidation capacity ( Figure 2B ), as seen previously after RYGB in the ZDF rat (12). The magnitude of fibrosis pathway downregulation was greater after RYGB-FMRR compared with RYGB. Using MCP-counter (54), SHAM-operated animals were predicted to have an increased relative abundance of renal cortical fibroblasts compared with SD controls ( Figure 2C ). Congruent with pathway analysis, both RYGB and RYGB-FMRR were predicted to decrease the relative abundance of renal cortical fibroblasts, with the magnitude of reduction being greater following RYGB-FMRR ( Figure 2C ).

Enrichment of fatty acid metabolism pathways was a unique and dominant transcriptomic response to RYGB-FMRR, which contrasted with decreased fatty acid metabolism, and in particular peroxisomal lipid metabolism, in SHAM-operated rats ( Figure 2B ). Peroxisomal and mitochondrial pathways upregulated by RYGB-FMRR are plotted to illustrate the abundance of peroxisome proliferator-activated receptor-alpha (PPARα)-responsive transcripts (for example, Acox1, Ehhadh, Acaa2) causing FAO pathway enrichment ( Figure 2D ). Gene ontology testing reinforced the unique stimulation of long-chain and very-long-chain fatty acid (VLCFA) metabolism, peroxisomal and mitochondrial activity, and fatty acid acyltransferase activity by RYGB-FMRR ( Supplemental Figures 3A–C ).

Activation of PPARα was predicted to increase FAO following RYGB-FMRR by upstream regulator analysis ( Figure 2E ). We validated renal expression of peroxisomal (Acox1, Ehhadh) and mitochondrial (Acaa2) PPARα-responsive transcripts by qRT-PCR ( Figure 2F ). PPARα-response genes were induced by RYGB-FMRR in the liver but not in visceral adipose tissue ( Supplemental Figure 4 ). Renal cortical induction of the mitochondrial enzyme pyruvate dehydrogenase kinase 4 (Pdk4), which is a sensitive transcriptional marker of increased FAO (79), was also validated by qRT-PCR.

We assessed cell type-specific expression patterns of transcripts in a human diabetic kidney single-nucleus RNA-sequencing dataset and in a rat tubular epithelial cell proteomics dataset (56, 58). Transcripts differentially regulated between the RYGB and RYGB-FMRR groups were most commonly expressed in the proximal tubule of the human and rat kidney ( Figures 3A, C ), and included PPARα-responsive genes with roles in peroxisomal (Acox1, Ehhadh) and mitochondrial (Acaa2) FAO ( Figures 3B, D ). We validated the proximal tubular enrichment of ACOX1 as well as its induction following RYGB-FMRR by immunohistochemistry ( Figure 3E ).

Using a network pharmacology approach, we assessed the magnitude, cellular localisation, and biological pathways associated with medication- and PPAR isotype-specific transcriptomic responses in rats treated with RYGB-FMRR compared with RYGB alone. Compared with other medications administered to RYGB-FMRR rats, there was a greater number of fenofibrate-responsive genes present in the RYGB-FMRR vs RYGB DEG list, both in absolute numbers (n=115 transcripts) and as a proportion of all genes known to be changed by fenofibrate (13%; Figure 4A ). Overall, 129 genes in the RYGB-FMRR vs RYGB DEG list were found to be responsive to one or more of the four medications in the FMRR combination, with 115 (89%) of these being fenofibrate-responsive ( Figure 4B ). Of the medication-responsive transcripts present in both the RYGB-FMRR vs RYGB DEG list and a rat tubular epithelial cell proteomics dataset (58), 86% of the fenofibrate-responsive transcripts were found to be proximal tubular-abundant, compared with only 40% of transcripts responsive to one or more of metformin, ramipril, or rosuvastatin, but not fenofibrate ( Figure 4C ).

Similarly, compared with other PPAR isotypes, there was a greater number of PPARα-responsive genes present in the RYGB-FMRR vs RYGB DEG list, both in absolute numbers (n=95 transcripts) and as a proportion of all genes known to be changed by PPARα (15%; Figure 4D ). Overall, 133 genes in the RYGB-FMRR vs RYGB DEG list were found to be responsive to one or more PPAR isotypes, with 95 (71%) of these being PPARα-responsive ( Figure 4E ). Of the PPAR isotype-responsive transcripts present in both the RYGB-FMRR vs RYGB DEG list and a rat tubular epithelial cell proteomics dataset (58), 86% of the PPARα-responsive transcripts were found to be proximal tubular-abundant, compared with only 52% of transcripts responsive to either PPARδ or PPARγ, but not PPARα ( Figure 4F ).

Furthermore, fenofibrate- and PPARα-responsive transcripts present in the RYGB-FMRR vs RYGB DEG list were confirmed to be functionally involved in stimulation of both peroxisomal and mitochondrial FAO by pathway over-representation analysis ( Figure 4G ). The magnitude of FAO pathway enrichment was similar between fenofibrate- and PPARα- responsive genes, thereby identifying PPARα as the principal mediator of fenofibrate-stimulated FAO. Thus, fenofibrate was found to be the dominant medication effector of gene expression changes following RYGB-FMRR, and via its molecular target PPARα, contributed to FAO induction in the proximal tubule. A network visualisation outlining the medication-responsive genes which are differentially expressed between the RYGB-FMRR and RYGB groups emphasises the dominant roles of fenofibrate and PPARα as effectors of gene expression changes following RYGB-FMRR ( Figure 5 ), which contributed to FAO induction in the proximal tubule.

Urinary metabolomic profiles of SHAM rats at baseline and follow-up clustered alongside baseline samples from RYGB and RYGB-FMRR rats, and were collectively designated as untreated ZDSD rats ( Figures 6A, B ). Untreated ZDSD rats separated into two subphenotypes, mild and severe, based on urinary metabolomic changes relating to disease severity. The major sources of variation along principal components 1 and 2 pertained to the untreated severe ZDSD rat phenotype and a post-RYGB phenotype, the latter of which was common to rats in both the RYGB and the RYGB-FMRR groups. Loading vectors in the principal component analysis biplot indicate that increased urinary excretion of sugars (glucose, sucrose, and mannose) and TCA cycle intermediates (citrate, succinate, and 2-oxoglutarate) characterised the untreated severe ZDSD rat phenotype ( Figure 6A ). Given that these rats clustered away from other untreated ZDSD rats on the basis of increased urinary excretion of sugars and TCA cycle intermediates, they were designated as having a more severe phenotype, while those with lower urinary abundance of these metabolites were designated as having a mild phenotype. Increased urinary excretion of host-gut microbial co-metabolites including N-phenylacetylglycine, 3-indoxyl sulfate, and hippurate occurred following both RYGB and RYGB-FMRR.

The RYGB and RYGB-FMRR groups clustered separately along principal components 3 and 4 ( Figure 6B ). The post-RYGB-FMRR urinary metabolome was characterised by increased abundance of PPARα-responsive metabolites involved in nicotinamide and vitamin B metabolism (1-methylnicotinamide, nicotinamide N-oxide, nicotinurate, and methylmalonate).

An RF model effectively classified the RYGB and RYGB-FMRR groups based on urinary metabolomic profiles, with an AUC of 0.97, sensitivity of 0.89, and specificity of 0.88. PPARα-responsive nicotinamide metabolites and TCA cycle intermediates were important to model performance ( Figure 6C ). Increased urinary excretion of PPARα-responsive nicotinamide metabolites occurred following RYGB-FMRR but not RYGB ( Figure 6D ). DKD-associated increases in urinary TCA cycle intermediates were generally reversed by both RYGB and RYGB-FMRR. Reductions in 2-oxoglutarate, fumarate, and malate were greater following RYGB-FMRR compared with RYGB. NMR characteristics of nicotinamide metabolites and TCA cycle intermediates are provided in Supplemental Table 1 .

Component 1 of a PLS-DA model integrating kidney RNA-Seq and urinary ¹H-NMR data selected a highly correlated (Pearson r = 0.93) set of transcripts (n=50) and metabolites (n=10) distinctive to RYGB-FMRR ( Figures 7A, B ). The majority of transcripts selected along component 1 have direct roles in renal cortical FAO, including peroxisomal (Ehhadh, Acox1) and mitochondrial (Acaa2) PPARα-responsive transcripts ( Figure 7C ). Component 1 metabolites included 1-methylnicotinamide, nicotinamide N-oxide, and 2-oxoglutarate. Network visualisation of the gene-metabolite signature distinctive to RYGB-FMRR illustrates positive correlations between FAO transcripts and urinary nicotinamide metabolites, as well as inverse correlations between FAO transcripts and urinary 2-oxoglutarate ( Figure 7D ).

Median [IQR] glomerular volume was elevated in SHAM-operated animals relative to the SD group (1.9x10⁶ [6.1x10⁵] vs 1.1x10⁶ [3.6x10⁵] μm³, p<0.001) ( Figures 8A, B ). Glomerular volume was reduced in RYGB-operated animals (p<0.001) and in animals treated with RYGB-FMRR relative to the SHAM group (p<0.001). Improvements in glomerular volume were greater in the RYGB-FMRR group compared with the RYGB group (1.3x10⁶ [4.6x10⁵] vs 1.4x10⁶ [5.0x10⁵] μm³, p=0.001).

Glomerular volume was inversely associated with induction of PPARα-responsive FAO transcripts including Acox1 (r=-0.47), Ehhadh (r=-0.47), and Acaa2 (r=-0.32) as well as urinary nicotinamide metabolites including 1-methylnicotinamide (r=-0.41) and nicotinamide N-oxide (r=-0.40), collectively suggesting that increased renal cortical PPARα activity following RYGB-FMRR was associated with quantitative improvements in glomerular structure ( Figure 8C ). Urinary excretion of TCA cycle intermediates was positively correlated with glomerulomegaly. Correlations with glomerular volume were moderately strong for two TCA cycle intermediates which were diminished to a greater degree by RYGB-FMRR than RYGB alone: 2-oxoglutarate (r=0.57) and fumarate (r=0.61).

Median [IQR] podocyte foot process frequency (PFPF) was reduced in SHAM-operated animals relative to the SD group (12.5 [4.0] vs 19.4 [4.3] FPs per 8μm GBM, p<0.001) ( Figures 9A, B ). Compared with the SHAM group, PFPF was partially restored by RYGB (15.0 [4.3] FPs, p=0.004) and by RYGB-FMRR (17.0 [3.5] FPs, p<0.001). Restoration of PFPF was greater in the RYGB-FMRR group compared with the RYGB group (p=0.04). Podocyte foot process diameter (PFPD) was higher in the SHAM group relative to the SD group (463 [238.1] vs 330.8 [135.6] nm, p<0.001) ( Figures 9A, C ). PFPD measures were significantly lower in both the RYGB (p=0.001) and RYGB-FMRR (p<0.001) groups relative to the SHAM group. Reduction in PFPD was greater in the RYGB-FMRR group compared with the RYGB group (344.8 [100.8] vs 410.9 [145.7] nm, p<0.001). Increases in GBM thickness were observed in the SHAM group relative to the SD group (330.5 [102.5] vs 219.4 [51.1] nm, p<0.001) ( Figures 9A, D ). GBM thickness was significantly lower in both the RYGB (p<0.001) and RYGB-FMRR (p<0.001) groups relative to the SHAM group. Improvements in GBM thickness were greater in the RYGB-FMRR group compared with the RYGB group (263.2 [59.2] vs 273.8 [63.5] nm, p=0.04).

Induction of FAO transcripts and urinary excretion of nicotinamide metabolites by RYGB-FMRR was inversely associated with glomerular ultrastructural injury (PFPD and GBM thickness) ( Figure 9E ). Urinary excretion of TCA cycle intermediates was positively correlated with glomerular ultrastructural injury (PFPD and GBM thickness).

Mitochondria in the pars convoluta had a greater two-dimensional area, were longer, and less round compared with mitochondria in the pars recta ( Supplemental Figure 5 ), as previously described (40). In the pars convoluta, median [IQR] mitochondrial roundness did not differ between the SD, SHAM, and RYGB groups at 0.36 [0.30], 0.37 [0.33], and 0.38 [0.30], respectively (p>0.05) ( Figures 10A, B ). However, mitochondrial roundness did decrease in the pars convoluta of animals in the RYGB-FMRR group at 0.35 [0.28] (p=0.02 versus SHAM and p=0.008 versus RYGB).

In the pars recta, mitochondrial roundness was increased in SHAM-operated animals relative to the SD group (0.69 [0.27] vs 0.65 [0.30], p<0.001) ( Figures 10C, D ). Mitochondrial roundness was lower in RYGB-operated animals relative to the SHAM group (0.68 [0.30] vs 0.69 [0.27], p=0.02). Mitochondrial roundness was lower in animals treated with RYGB-FMRR at 0.62 [0.32] (p<0.001 versus both SHAM and RYGB).

Increased renal FAO transcript expression and urinary nicotinamide metabolite excretion following RYGB-FMRR inversely correlated with mitochondrial roundness in the proximal tubule ( Figures 10E ), with correlations being greater in magnitude in the pars recta. Urinary excretion of TCA cycle intermediates positively correlated with proximal tubular mitochondrial roundness.

Enhanced expression of PPARα-responsive transcripts and reduced mitochondrial roundening may have been related to greater improvements in metabolic control after RYGB-FMRR compared with RYGB. We therefore assessed FAO transcript expression and mitochondrial roundness in a pair of rats, one each from the RYGB and RYGB-FMRR groups, that were matched for baseline values and delta improvements in body weight, plasma glucose, and albuminuria. Both animals achieved >20% reduction in body weight, >40% reduction in plasma glucose, and >25% reduction in albuminuria ( Figure 11A ). Despite this, expression of FAO transcripts was higher in the RYGB-FMRR animal ( Figure 11B ). Furthermore, mitochondrial roundness of the RYGB-FMRR animal was lower compared with the matched RYGB animal in both the pars convoluta (0.36 [0.29] vs 0.43 [0.32], p=0.04) and the pars recta (0.56 [0.35] vs 0.64 [0.25], p <0.001) ( Figure 11C ).