The primary antibodies AMPKα, p-AMPKα (Thr172), insulin receptor substrate (IRS), p-IRS (Ser307), phosphatidylinositol 3-kinases (PI3K), p-PI3Kp85 (Tyr458)/p55 (Tyr199), protein kinase B (AKT), p-AKT(Ser473), acetyl-CoA carboxylase (ACC), p-ACC (Ser79), thioredoxin-interacting protein (TXNIP) and the second antibodies anti-mouse IgG and anti-rabbit IgG were purchased form the Cell Signaling Technology Inc. (Boston, MA, United States). The primary antibodies peroxisome proliferator-activated receptor-alpha (PPAR-α) and PPAR-β were purchased from the Santa Cruz Biotechnology Co., Ltd. (Dallas, United States) and carnitine palmitoyltransferase-1 alpha (CPT-1α) was obtained from the Abcam (Cambridge, United Kingdom). POD (purity above 95%) and RES (purity above 99%) were obtained from the Aladdin Biochemical Technology Co., Ltd. (Shanghai, China). Fructose, valeric acid sodium and caproic acid sodium were purchased from the Maclin Biochemical Technology Co., Ltd. (Shanghai, China). Recombinant human insulin was bought from Tonghua Dongbao Pharmaceutical Co., Ltd. (Tonghua, China). N-acetylcysteine (NAC) was purchased from the Sigma-Aldrich (St. Louis, MO, United States). Other chemicals and drugs were of the highest grade available.

The isoform and binding site of AMPK (α1β1γ1) was confirmed according to the previously reported AMPK direct activator A-769662 (43–45). The three-dimensional (3D) structure of AMPK was downloaded from the Protein Data Bank (ID: 4ZHX), and the ligand 2D and 3D structures of POD and RES were constructed by Chem3D Ultra (Version 8.0), respectively. The ligand (POD or RES) was docked into the active site of the prepared AMPK crystal structure by Schrodinger (Version 12.5). The binding ability of POD or RES with AMPK was evaluated by docking scores.

Male C57BL/6J mice, 5 weeks old and weighing 20 ± 2 g, were purchased from the Guangdong Provincial Laboratory Animal Center Co., Ltd. (Guangzhou, China). Mice were housed in a room with controlled temperature (23 ± 2°C, 40 ± 10% relative humidity, and 12-h light–dark cycle), and were allowed free access to water ad libitum and food.

Experiment 1: The animal model of non-alcoholic fatty liver disease was induced by 10% (w/v) fructose for 3 weeks and 30% fructose for 5 weeks in drinking water. The mice were randomly divided into four groups based on body weight (n = 5–8): Mice in Control group were given ad libitum access to drinking water, mice in Model group were given 30% fructose-containing drinking water, and mice in POD or RES group were given 30% fructose-containing drinking water and 50 mg/kg/day POD or RES by oral gavage, respectively, for another 10 weeks (Figure 2). After that all mice were fasted overnight, peripheral blood was collected from the ophthalmic vein after anesthetized and then the mice were sacrificed by cervical dislocation. Serum was obtained by centrifugation (5,000 round/min, 10 min, 4°C) and stored at −80°C. Livers were collected and stored at −80°C. The fresh fecal samples of each mouse were collected and stored at −80°C during the final week of the animal experiment. POD and RES (5 mg/mL) were suspended sufficiently in the carboxymethylcellulose sodium aqueous solution (0.5%, w/v).

Experiment 2: The animal model of non-alcoholic fatty liver disease was induced by fructose (30%, w/v) in drinking water for 5 weeks. The mice were randomly divided into eight groups based on the body weight (n = 5–7): Mice in Control group were given ad libitum access to drinking water, mice in Model group were given 30% fructose-containing drinking water, mice in POD group were given 30% fructose-containing drinking water and 25 mg/kg/day POD by oral gavage, mice in valeric acid sodium group were given 30% fructose and 100 mmol/L valeric acid sodium-containing drinking water, mice in caproic acid sodium group were given 30% fructose and 100 mmol/L caproic acid sodium-containing drinking water, mice in POD plus Valeric acid sodium group were given 30% fructose and 100 mmol/L valeric acid sodium-containing drinking water and 25 mg/kg/day POD by oral gavage, mice in POD plus caproic acid sodium group were given 30% fructose and 100 mmol/L caproic acid sodium-containing drinking water and 25 mg/kg/day POD by oral gavage, and the mice in NAC group were given 30% fructose-containing drinking water and 0.5 g/kg/day NAC by oral gavage, for another 6 weeks (Figure 2). After that all mice were fasted overnight and sacrificed by cervical dislocation after anesthetized. Liver and serum were harvested and stored at −80°C. The drinking fluid was exchanged daily.

One touch glucometer (Roche Diagnostics, Mannheim, Germany) was used to measure the fasting blood glucose (FBG) levels on tail vein blood of mice after overnight fasting. For the insulin tolerance tests (ITT), after overnight fasting, insulin (0.3 unit) was administered by the intraperitoneal injection, and the blood glucose was measured at 0.5, 1.0, 1.5, and 2.0 h after the injection with the glucometer.

Serum aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine (Cr), and blood urea nitrogen (BUN) levels were measured with an automatic biochemical analyzer (HITACHI 7020, Japan). Kits for triglyceride (TG), total cholesterol (TC), free fatty acid (FFA), LDL-cholesterol (LDL-C), MDA, and SOD obtained from Nanjing Jiancheng Bioengineering Institute (Nanjing, China). The levels of serum insulin and HbA1c were measured using the commercial ELISA kits from Jiangsu Meimian industrial Co., Ltd. (Yancheng, China) and Jiangsu Boshen Biotechnology Co., Ltd. (Nanjing, China). Homeostasis model assessment-insulin resistance (HOMA-IR) index was calculated by the following formula:

Livers (0.1 g) were homogenized in ice-cold tissue extraction buffer (1 ml) and then the lysates were clarified by centrifugation (12,000 g, 4°C, 15 min), after that the supernatant was collected for analysis. The levels of TC, TG, and FFA were measured according the corresponding protocols in the commercial kits provided by manufacturers.

Harvested liver specimens were fixed in paraformaldehyde solution at room temperature, dehydrated, embedded in paraffin, after that a fully automated rotary microtome (LEICA RM2255, Shanghai, China) was used to cut the paraffin into 5 μm thickness serially. Periodic acid-schiff (PAS), hematoxylin-eosin (HE), MASSON, and oil red O stains were performed according to the standard protocols of the corresponding commercial kits, respectively, and then the slices were pictured with electron microscope (Nikon eclipse ti, Japan). All the assays were performed in a blinded manner.

Total RNA was extracted from liver tissues of mouse using TRIzol reagent, obtained from Takara Biotechnology, according to the manufacturer’s protocol. The cDNA was prepared using 50 ng of total RNA by the reverse transcription according to the manufacturer’s instructions. SYBR Green qPCR SuperMix was performed on a CFX System (Bio-Rad, Hercules, CA, United States) according to the manufacturer’s instructions. Real-time PCR of cDNA was performed using standard PCR cycling condition. Relative expression level of target gene was normalized against control group β-actin and presented as a ratio to the expression level in other groups with the formula 2^–(ΔΔCt). The primer sequence of each tested gene is shown in Supplementary Table 1.

Livers (0.1 g) were homogenized in 0.1 mg/ml phenylmethylsulfonyl fluoride-containing ice-cold RIPA buffer (1 ml) and then the lysates were clarified by centrifugation (12,000 g, 4°C, 15 min), after that the supernatant was collected and stored at −80°C. The samples were denatured with loading buffer (99°C, 10 min), then the equal protein were separated by SDS-PAGE gel electrophoresis and transferred onto a polyvinylidene fluoride (PVDF) membrane. The membrane was probed with primary antibody according to the dilution ratio provided by manufacturers overnight at 4°C, and then incubated with secondary antibody according to the dilution ratio provided by manufacturers at room temperature for 60 min. The immunoreactivity was detected using the ChemiDoc XRS + detection system (ECL, Bio-Rad, United States). The densitometric analysis was performed with Quantity One^® Image Analyzer software program (Bio-Rad). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used for normalization.

Total genome DNA of bacterial was extracted from frozen feces with QIAamp DNA stool Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s guideline. The specific primer with the barcode (16S V3 + V4) was used to amplify the 16S rDNA gene. TruSeq^® DNA PCR-Free Sample Preparation Kit (Suzhou RENOLD Biological Technology Co., Ltd., Suzhou, China) was used to construct the DNA sequencing libraries. Fast Hifidelity Polymerase and Phusion^® High-Fidelity PCR Master Mix with GC Buffer (New England Biolabs Co., Ltd., Beijing, China) were used for The PCR amplification under the standard thermal cycling and extension conditions. Paired-end sequencing of the PCR products was performed on the NovaSeq6000 at Suzhou Bionovogene Co., Ltd. (Suzhou, China).

Gas chromatography-mass spectrometer method (Thermo TRACE 1310-ISQ LT instrument, Agilent HP-INNOWAX column with 30 m × 0.25 mm ID × 0.25 μm particle size) (Suzhou Bionovogene Co., Ltd., China) was used to measure the fecal levels of acetic acid, propionic acid, butyrate, isobutyric acid, caproic acid, valerate and isovaleric acid. 50 μL phosphoric acid (15%), 100 μL internal standard (isohexic acid) solution (125 μg/mL) and 400 μL ether were homogenated with 50 mg fresh feces for 1 min, then the mix were centrifuged at 12000 rpm and 4°C for 10 min. After that the supernatant was collected for measurement.

Data were expressed as mean ± SEM, and compared using the Student’s t-test or two-way ANOVA post hoc Bonferroni test, as appropriate (GraphPad Prism 5 Software, Inc., La Jolla, CA, United States). The correlation coefficient between the fecal levels of SCFAs and the relative abundance of gut microbiota at the genus level were performed with Pearson correlation analysis (SPSS software, version 20, IBM, Armonk, NY, United States). The significance levels were established at a p-value of <0.05.

To better understand the interactions of POD and RES with AMPK, molecular docking study was performed to analyze the binding abilities of POD and RES with AMPK. The molecular docking results between the ligands (POD and RES) and the receptor (AMPK) suggest POD displayed a higher predicted binding score than RES (−8.677 vs. −6.802) (Figures 1B,E). This indicates that POD possesses a stronger interaction with AMPK than RES. Conventional hydrogen bond and van der Waals are the mainly binding bonds between POD and RES with AMPK. POD exerts higher binding ability with AMPK than RES may involve there are more hydroxyl groups in POD (Figure 1) because hydroxyl groups formed H-bonds to AMPK (Figure 1C) are important for stable binding. The more residues of AMPK units or binding pocket involved in the H-bond interaction or the binding of POD and AMPK by van der Waals also contributed to its binding (Figure 1C).

Body weight (Supplementary Figure 1A) and serum parameters, such as ALT, AST, Cr, and BUN (Supplementary Table 2), showed that the doses of POD and RES employed did not possess toxicity in mice. Since the intake of extra calories from fructose in drinking fluid (Supplementary Figure 1B), the food intake was lowered (Figure 3A). Both the POD and RES administrations did not significantly influence the fluid intake (Supplementary Figure 1A), but, interestingly, elevated the food intake of the mice (Figure 3A). High-fructose diet enhanced the insulin tolerance (Figures 3B,C), elevated the levels of serum insulin, FBG, and serum HbA1c (Figures 3D–F) and caused the hepatic glycogen accumulation and fibrogenesis (Figures 3G,H). POD and RES alleviated these lesions with a similar effect (Figures 3D–H). In addition, serious lipid accumulation (Figure 3I), fatty infiltration and amyloidosis (Figure 3J) of liver were also observed in the model mice. POD showed a better effect on preventing these lesions than RES (Figures 3I,J). This was also validated by the higher capacity of POD in reducing the levels of TC and TG in liver (Figures 3K,L) and serum (Figures 3M,N). These results suggest that POD and RES can alleviate the glucolipid dysmetabolism, and POD is superior to RES on ameliorating non-alcoholic fatty liver disease than RES in mice fed a high-fructose diet.

At the molecular level, fructose consumption suppressed the AMPK signaling pathway, such as downregulation of p-AMPK (Thr172), CPT-1α, PPAR-α, PPAR-γ, p-ACC (Ser79), p-AKT (Ser473), -PI3Kp85 (Tyr485), and p-IRS1 (Ser307) and upregulation of TXNIP, in liver (Figures 4A,B). POD significantly prevented these alterations (Figures 4A,B), and RES displayed a preventing effect on the downregulation of p-AMPKα (Thr172) (Figure 4A), but without affecting the levels of the downstream proteins PPAR-γ, p-ACC (Ser79), p-AKT (Ser473), and p-PI3Kp85 (Tyr485) (Figures 4A,B). In addition, long-term fructose consumption caused elevation of oxidative stress level in liver indicated by the pronouncedly increase of TXNIP (Figure 4A) and MDA (Figure 4E) and decrease of SOD (Figure 4F), both POD and RES administrations effectively prevented these alterations (Figures 4E,F). However, neither POD nor RES affected the mRNA levels of GPR41 and GPR43 in liver (Figures 4C,D).

The α-diversity analysis of the gut microbiomes showed frcutose consumption lowered the diversity of microbiota (Figures 5A,B). POD and RES administrations markedly increased the microbiota diversity evidenced by the elevated Shannon and Simpson indexes (Figures 5A,B). The β-diversity analysis evaluated the overall differences in groups, results showed a distinct clustering of gut microbial community structure in control and model groups (Figure 5C), POD and RES administrations altered the structure in a similar trend (Figure 5C). The Venn diagrams were used to show the over lapping operational taxonomic units (OTUs), which displayed the similarity and consistency of samples. There are 313 OTUs shared in all groups (Figure 5D). Three hundred and 68 OTUs were identified in control group, and fructose consumption increased the OTUs numbers to 509 (Figure 5D). Interestingly, POD administration lowered the OTUs numbers to 398 and RES increased the OTUs numbers to 511 (Figure 5D). These results indicated long-term intake of fructose markedly affected the community structure and relative abundance of gut microbiomes (Figures 5A–E), and POD has a higher regulating effect on gut microbiomes than RES (Figure 5D). The function of the significantly altered gut microbiota (Figures 5F–K) post POD or RES treatment will be discussed in the discussion section.

Our results showed the levels of fecal SCFAs of mice in the model group without a significant alteration compared to control (Figures 5L–N), POD treatment markedly elevated the levels of valeric acid and caproic acid (Figures 5M,N), but RES did not pronouncedly affect the levels of SCFAs (Figures 5L–N). Correlation analyses between SCFAs levels and relative abundance of gut microbiota indicated the valeric acid level was significantly negative correlated with the abundance of Parasutterella (Figure 5O) and positive correlated with the abundance of Acetitomaculum, Anaerotruncus, Candidatus_Soleaferrea, Colidextribacter, Harryflintia, Mucispirillum, Negativibacillus, Butyricimonas, [Eubacterium]_ventriosum_group, Candidatus_Soleaferrea, Bifidobacterium, Monoglobus, Lactococcus, Allobaculum, Intestinimonas, Lachnoclostridium, [Eubacterium]_nodatum_group, Streptococcus, Romboutsia, Anaerovorax, and unidentified_Ruminococcace at the genus level (Figure 5O). However, the caproic acid level was significantly negative correlated with the abundance of Escherichia-Shigella, NK4A214_group, Roseburia and Colidextribacter, and positive correlated with the abundance of Anaerofustis and [Eubacterium]_ventriosum_group (Figure 5O).

The markedly elevated levels of fecal valeric acid and caproic acid via modifying gut microbiomes by POD administration may be the main reason that POD exerted a higher property on ameliorating lipid dysmetabolism than RES. Thereby, we next investigated the regulating effects of valeric acid and caproic acid on AMPK and lipid metabolism in mice fed a high-fructose diet. Results showed that both caproic acid sodium and valeric acid sodium reduced the lipid accumulation, and the levels of TG and FFA in liver (Figures 6A,C,D). In addition, caproic acid sodium decreased the levels of TC, TG, FFA, and LDL-C in serum (Figures 6E–H). Importantly, we found when the low dose of POD (25 mg/kg) without a pronouncedly improvement on non-alcoholic fatty liver disease, caproic acid sodium and/or valeric acid sodium can enhance the effects of POD on preventing lipid dysmetabolism (Figures 6A,B,F–H). Consistently with these improving effects of SCFAs (caproic acid sodium and valeric acid sodium) or SCFA plus POD on lipid metabolism, AMPK was activated post the treatment indicated by the significantly upregulation of p-AMPK α (Thr172) and p-ACC (Ser79) (Figure 7). Neither caproic acid sodium nor valeric acid sodium affected the mRNA levels of GPR41 and GPR43 in liver.