Male C57BL/6 mice were purchased from The Hong Kong Polytechnic University. The animal study and treatment procedures were approved by our Animal Ethics Committee. All mice were housed at a temperature of 25 ± 2°C and humidity of 60 ± 5% with a 12 h light and dark cycle and were given ad libitum access to water and diet. The 8-week-old mice were divided randomly into three subgroups: standard chow control (STC), high-fat diet-induced diabetic nephropathy (HFD), and high-fat diet-induced diabetic nephropathy with dulaglutide treatment (H + Dula). The STC group was fed a standard chow (13.2% fat, 24.7% protein, and 62.1% carbohydrates; PicoLab Rodent Diet 20 #5053, St. Louis, United States), and others were fed with a high-fat diet (60% fat, 20% protein, and 20% carbohydrates; Research Diets Inc., New Brunswick, United States) for 12 weeks. Then, the treatment group was intraperitoneally administrated with dulaglutide in the morning at a dose of 0.6 mg/kg body weight (BW) once every 7 days and lasting for 4 weeks. An equal volume of saline was also injected into mice in the STC and HFD groups.

The full body composition of mice was analyzed using a Minispec LF-50 nuclear magnetic resonance spectrometer according to the manufacturer’s protocol (Bruker, Germany). The blood glucose of the mice was measured from the tail vein using the Accu-Chek^® glucometer (Roche Diagnostics, Indiana, United States). The 24 h urine volume was measured and collected before week 19 and after week 24 of dulaglutide treatment by individual metabolic cages. Urine samples were centrifuged at 2,000 × g for 10 min at 4°C, aliquoted, and stored at −80°C before analysis. Albumin concentration was determined using the mouse albumin ELISA (ab207620, Abcam, United Kingdom) kit according to the manufacturer’s instructions. Serum urea level was determined using the QuantiChrom Urea Assay Kit (BioAssay Systems, United States), and blood urea nitrogen (BUN) level was calculated. For intraperitoneal glucose tolerance test (IPGTT), the mice were then fasted overnight for 16 h with ad libitum access to water only. Basal tail vein fasting blood glucose was obtained and measured before intraperitoneal injection of glucose solution at a dose of 1.5 g/kg/mouse. After the completion of high-fat diet feeding and dulaglutide treatment, mice were fasted overnight for 16 h and were anesthetized for blood collection by cardiac puncture. Metabolic organs including the kidneys and liver were harvested and separated for snap-freezing in liquid nitrogen or formalin-fixed for subsequent analyses.

Tissues were isolated and fixed for 24 h in 10% neutral-buffered formalin for subsequent tissue processing and embedded in paraffin (FFPE). The FFPE sections of 4 μm thickness were prepared and mounted on uncoated glass slides for hematoxylin and eosin (H&E) and periodic acid Schiff–methenamine silver (PASM) staining. Oil Red O (ORO) staining for neutral lipid was performed on snap-frozen tissues. Tissues were trimmed and cryosectioned using a CryoStar NX70 cryostat (Thermo Scientific, United States) at 8 μm thickness and then thaw-mounted on uncoated glass slides. They were dried for 5 min and fixed in 4% neutral-buffered formalin before proceeding with ORO staining according to the manufacturer’s protocol (BioGnost, Croatia). Additional kidney cryosections were thaw-mounted on indium tin oxide (ITO) slides (Bruker, Germany). The slides were dried under vacuum in a desiccator for approximately 30 min and stored at −80°C until use within 7 days.

The tissue samples were thawed for 30 min at 4°C. Lipids were extracted according to Bligh and Dyer’s (1959) method. Briefly, 200 μl of ice-cold methanol was added to each sample, vortexed for 15 s, and sonicated in a pre-cooled sonicator for 15 min. Then, 200 and 100 μl of ice-cold chloroform and water were added, respectively, into each sample, vortexed for 15 s, and sonicated for another 15 min. The resulting mixture was centrifuged at 10,000 RPM for 5 min at 4°C. The lower phase containing lipids was separated and dried using a SpeedVac concentrator (Labconco, MO, United States) prior to storage at −80°C. Before Ultra-high-performance liquid chromatography/electrospray ionization-quadrupole time-of-flight mass spectrometry (UHPLC/ESI-QTOF-MS) analysis, the lipid fractions were reconstituted in 100 μl of ice-cold methanol and chloroform (1:1, v/v), followed by sonication for 15 min. Samples containing 10 μl of lipids were pooled and vortexed to form the quality control (QC) samples.

UHPLC/ESI-QTOF-MS analysis was performed in both positive and negative ionization modes with an Agilent 6540B ESI-QTOF-MS connected with an Agilent 1290 UHPLC (Agilent, United States). Liquid chromatography was carried out using a Waters ACQUITY BEH C18 column. The mobile-phase solvents consisted of A (60:40 acetonitrile:water (v/v) with 5 mM ammonium formate and 0.1% (v/v) formic acid) and B (90:10 isopropanol:acetonitrile (v/v) with 5 mM ammonium formate and 0.1% (v/v) formic acid). Chromatographic separation was performed at a flow rate of 0.3 ml/min with an elution gradient. Column and sample chamber temperatures were set at 55°C and 4°C, respectively.

Raw LCMS data collected were subjected to peak picking, alignment, and normalization (by all compound abundance) by the Progenesis QI software (Nonlinear Dynamics-Waters, Milford, United States), from which a dataset consisting of retention time, mass-to-charge ratio (m/z), and normalized abundance was generated. The dataset was further processed by the EZInfo 2.0 software (Sartorius, Germany) for multivariate data analysis, including principal component analysis (PCA) and orthogonal partial least-squared discriminant analysis (OPLS-DA). OPLS-DA was used to create the variable importance in projection (VIP) plot. Compounds with a VIP value >1 and t-test with p < 0.05 were considered as potential differential markers, which were identified with the Human Metabolome database and LIPID MAPS Structure database (LIPID MAPS Lipidomics Gateway) based on accurate molecular mass and fragment ion pattern.

A matrix of 10 mg/ml of 2,5-dihydroxybenzoic acid (DHB) was evenly sprayed onto the ITO slide using an HTX TM-Sprayer (HTX Imaging, North Carolina, United States). The homogeneity of the matrix layer was assessed using a light inverted microscope. A peptide calibration standard (Bruker, Germany) was then applied to the slide for mass calibration. Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) analysis was performed with a Bruker UltrafleXtreme matrix-assisted laser desorption/ionization time-of-flight/time-of-flight mass spectrometer equipped with a 355 nm smart beam laser source (Bruker, Germany). MALDI-MSI data were acquired with the Bruker flexImaging program. The spatial resolution was set at 100 × 100 μm, and mass spectra were collected with a m/z range of 400–2,000 and 500 consecutive laser shots accumulated per pixel. Data analysis was performed with the SCiLS Lab software version 2020b Pro (Bruker, Germany). Root-mean-square normalization was applied to normalize the ion intensity during ion image generation. Lipids were identified with the LIPID MAPS (LIPID MAPS Lipidomics Gateway) database based on accurate molecular mass and fragment ion pattern.

Total RNAs were extracted from kidney tissue samples using RNAiso Plus (Takara, China) according to manufacturer’s instruction and were quantified using a NanoDrop One spectrophotometer (Thermo Scientific, United States). First-strand complementary DNA (cDNA) synthesis was carried out from 1 μg RNA using PrimeScript RT Kit (Takara, China). TB Green Premix Ex Taq was incorporated with cDNA and primers for subsequent amplification and fluorescent detection. Amplification was performed using LightCycler^® 96 (Roche, Germany). Relative gene expression was calculated using the 2^−ΔΔCt method. Supplementary Table S1 shows the primer sequences used in this study.

Statistical analysis was performed routinely using GraphPad Prism 8 (GraphPad Software, San Diego, CA), with p < 0.05 being considered as significant. All values were expressed as mean ± standard error of the mean (SEM). For comparisons among groups, one-way ANOVA was performed with Tukey post-hoc test with the HFD group set as the reference group to compare multiple groups. Each experiment was repeated at least three times.

Previous studies demonstrated that GLP-1RAs could be utilized in treating DKD in humans (Marso et al., 2016a; Marso et al., 2016b; Muskiet et al., 2018; Tuttle et al., 2018; Gerstein et al., 2019). To explore the mechanism, we used high-fat diet to promote renal injury in mice (Figure 1A). As previously reported (Findeisen et al., 2019; Park et al., 2019; Svendsen et al., 2020; Hupa-Breier et al., 2021), treatment of dulaglutide lowered the body weight (Figure 1B) by reducing the fat mass (Figure 1C and Supplementary Figures S1A−F), improved glucose homeostasis (Figures 1D–G), and reduced the energy expenditure (Figure 1H), respiratory exchange ratio (Figure 1I), and food intake (Supplementary Figure S1G). Importantly, dulaglutide improved kidney function in the H + Dula group mice and reduced BUN level (Figure 1J) and albuminuria (Figure 1K) compared with the HFD group.

We also performed histological examination and revealed that the kidney of high-fat diet-fed mice developed vacuolization of renal tubules, same as that observed in obese patients with DKD (Figure 2A, black arrow) (Liptak and Ivanyi, 2006; Watanabe et al., 2021). Reduction in the area of vacuolization was observed in the H + Dula group as compared to the HFD group (Figure 2A). PASM staining showed mesangial expansion, enlargement of the glomerular tuft, and thickening of the tubular basement membrane in the HFD group as compared with the STC group, and the area showed a significant reduction in size after dulaglutide treatment (Figure 2B).

To visualize the renal neutral lipid, ORO staining was used. A dramatic increase in lipids per cell in the HFD group as compared to the STC control was observed mainly in the cortex region of the kidney. The H + Dula group showed a significant reduction in the amount of lipid droplet per cell as compared with HFD group, but the lipid level was still much higher than that of the STC group (Figure 2C). High-fat diet also increased lipid accumulation within the medulla region but not as predominate as in the cortex region (Figure 2D). To assess for renal tubular injury in the kidney, immunohistochemical staining (IHC) was performed with kidney injury molecule-1 (KIM-1; Figure 2E). KIM-1 is highly expressed in renal tubules in the HFD group as compared with the STC group (p < 0.0001). A significant reduction of KIM-1 expression in the H + Dula group is found, consistent with previous urinary albumin data, indicating reduced damage in renal tubules after dulaglutide treatment. These results suggest that dulaglutide treatment attenuated the progression of high-fat diet-induced kidney damage in our mouse model. As previous study reported renal tubular GLP-1R expression is reduced in chronic kidney disease (Choi et al., 2019), IHC staining of kidney sections was performed to explore GLP-1R protein expression. In brief, there were no significant changes in GLP-1R expression after high-fat diet and dulaglutide treatment in the renal tubules (Figure 2F).

To explore the changes in lipidomes of kidney in high-fat diet-fed mice after dulaglutide treatment, UHPLC/ESI-QTOF-MS analysis was performed. The general visualization of interrelations and separation among the STC control, HFD, and H + Dula groups was plotted (Figure 3). PCA score plots show that the lipidomic profiles of the HFD and H + Dula groups are clustered and separated from the STC control group (Figures 3A,B). To further facilitate the detection and differentiation of potential compounds of interest among the three groups, PLS-DA was plotted. PLS-DA shows a clear separation among all three groups (Figures 3C,D). In addition, an OPLS-DA analysis was performed on mass spectra and normalized abundance obtained from the HFD and H + Dula groups to show a significant separation between these two groups (Figures 3E,F). These results suggest that 4 week dulaglutide treatment dramatically changed the lipid species in the kidneys of high-fat diet-fed mice.

Pathway analysis was performed and identified several altered metabolic pathways in the kidneys (Figure 3G). The top altered pathway is glycerophospholipid metabolism. Previous studies demonstrated that the composition and abundance of glycerophospholipids are changed in the serum and muscle of obese mice with T2DM (Rauschert et al., 2016; Wolfgang, 2021). As glycerophospholipids can undergo esterification with amino acids to form a range of glycerophospholipid subtypes, unbiased lipid set enrichment analysis was performed to screen for differently expressed lipids. Intriguingly, an essential constituent of the inner mitochondrial membrane lipid, cardiolipin, shows the highest enrichment (Figure 3H). As the above analyses did not provide any information for their relative levels, we further analyzed the changes in the level of each lipid species in the HFD and H + Dula groups. In short, we identified 62 lipids of interest and listed their details in Table 1. In brief, there was reduced abundance of most identified species for diacylglycerols (DG), phosphatidic acids (PA), phosphatidylglycerols (PG), and triglycerides (TG) in the kidney of the H + Dula group. Unexpectedly, the levels of a number of lipid species were remarkably increased in the H + Dula group, such as all identified ceramides (Cer), four out of five cardiolipin species (CL), and eight out of nine phosphatidylethanolamine species (PE).

To explore the changes in the lipid species distribution, kidney cross sections were examined by MALDI-MSI. Consistent with our ORO staining and UHPLC/ESI-QTOF-MS results, high abundance of TG is observed, especially in the kidney cortex of high-fat diet-fed mice. For example, a low abundance of TG (52:3) (m/z = 895.8) was found in the outer medulla of the cortex region of the STC control by MALDI-MSI (Figure 4A, left panel). After high-fat diet treatment, TG (52:3) abundance increased dramatically in the cortex region and outer medulla of the HFD group (Figure 4A, middle panel). Decreased signal intensities of TG (52:3) due to dulaglutide treatment were observed mainly in the cortex region (Figure 4A, right panel).

Unexpectedly, a number of lipid species such as PI (38:4) (m/z = 885.7) and PC (38:7) (m/z = 804.6) decreased in abundance after the high-fat diet treatment (Figures 4B,C, left vs middle panels), and dulaglutide treatment almost restored their levels in the kidney cortex of the HFD group to the same level as that of STC group (Figures 4B,C). For cardiolipin, CL (74:2) (m/z = 1,488.2) was used as an example. A high abundance of CL (74:2) in the cortex region of mouse kidney of the STC control was observed as compared with that of the HFD group (Figure 4D, left vs middle panels). The reduction of CL (74:2) in the HFD group was ameliorated after dulaglutide treatment, especially for the cortex region (Figure 4D, middle vs right panels). In summary, both UHPLC/ESI-QTOF-MS and MALDI-MSI showed an induction in the abundance of cardiolipin, especially within the kidney cortex regions of the HFD group after dulaglutide treatment.

As cardiolipin is a unique inner mitochondrial membrane lipid (Paradies et al., 2014), to explore whether the level of cardiolipin was associated with the number of mitochondria, we checked their amount of mitochondrial DNA (mtDNA) by RT-qPCR. In agreement with their cardiolipin levels, the STC group had the highest copy number of mtDNA (Figure 5A). High-fat diet treatment lowered the copy number of mtDNA in the kidney samples, while treatment of dulaglutide increased it (Figure 5A).

Previous studies demonstrated that GLP-1 increases mitochondrial biogenesis in various cell types via the expression level of the transcriptional coactivator PGC-1α (Li et al., 2011; Quan et al., 2020). We checked the expression level of PGC-1α and its regulated genes. Dulaglutide treatment increased the mRNA expression of PGC-1α-regulated genes NRF1 by only ∼30% (Figure 5C) and TFAM by only ∼50% (Figure 5D), but only ∼15% of mRNA expression of the PGC-1α (Figure 5B). There was no induction in the mRNA expression of PGC-1α–dependent gluconeogenic gene PEPCK (Figure 5E). The levels of mRNA expression of mitochondrial biogenesis genes ND5 complex I (Figure 5F), CYTB (complex III, Figure 5G), and COX III (complex IV, Figure 5H) were also unchanged. We further measured the protein expression of PGC-1α, and there was no change (Supplementary Figure S3). Thus, it is conceivable that dulaglutide’s increasing of the mtDNA copy numbers is not via mitochondrial biogenesis.

As dulaglutide decreased cardiolipin precursors phosphatidic acid and phosphatidylglycerol levels (Table 1), we explored whether dulaglutide induced cardiolipin via the cardiolipin synthesis pathway (Figure 6A). We measured the mRNA expression of key cardiolipin synthesis enzymes (CDS1, PGPS, CLS, and TAZ). In brief, the mRNA expression of all four enzymes in H + Dula showed a significant increase as compared with that in the STC and HFD groups (Figures 6B–E). As GLP-1R is not expressed in hepatocytes (Pyke et al., 2014; Richards et al., 2014), we also checked the lipid accumulation and the expression of cardiolipin synthesis genes in their livers by ORO staining and RT-qPCR. Consistent with recent findings (Hupa-Breier et al., 2021), dulaglutide treatment cannot lower the hepatic lipid level of high-fat diet-fed mice (Figures 7A,B), and the expression level of hepatic cardiolipin synthesis genes remained unchanged as compared with HFD and H + Dula groups (Figures 7C–F). In summary, our data suggested that dulaglutide induces the expression of major enzymes in the cardiolipin synthesis pathway and hence enhances cardiolipin level in the kidney cortex of high-fed-diet-fed mice.