Chemoproteomics reveals baicalin activates hepatic CPT1 to ameliorate diet-induced obesity and hepatic steatosis

To facilitate our quantitative chemical proteomic study to identify the targets of baicalin, we first evaluated the ability of baicalin to lower the lipid accumulation in a mammalian cell line. HeLa cells were treated with 1 mM oleic acid and palmitic acid (2:1 molar ratio) to induce deposition of lipid droplets. These cells were then treated with a series of concentrations of baicalin for 24 h in presence of the excessive free fatty acids. The cells were stained by Oil Red O (ORO), and the extent of lipid accumulation was quantified by the intensity of DMSO-extracted ORO dye. Baicalin is able to significantly lower the cellular accumulation of lipid droplets at as low as 12.5 μM and maintains its strong lipid-reducing effect at up to 400 μM when ∼80% of the cells are still viable (Fig. 1B). Baicalin treatment at 100 μM can already reduce the amount of cellular lipids by ∼50% (Fig. 1C). Similar results were obtained when ORO-stained cells were directly imaged and quantified (Fig. 1D and SI Appendix, Fig. S1A). Furthermore, we monitored the lipid content in these cells by stimulated Raman scattering (SRS) microscopy (31). A dramatic decrease of SRS signals in the baicalin-treated cells confirmed baicalin’s activity in reducing lipid accumulations in HeLa cells (Fig. 1E and SI Appendix, Fig. S1B). In addition, we isolated hepatocytes from mouse liver and confirmed with ORO staining that baicalin retains similar activity in lowering lipid accumulation in primary cells (SI Appendix, Fig. S2A).

We next employed activity-based protein profiling (ABPP) to identify the protein targets that directly interact with baicalin in proteomes. The ABPP-based chemical proteomic strategy has enabled discovery of functionally unannotated enzymes and deconvolution of protein targets of bioactive small molecules (32, 33). In particular, its combination with stable isotope labeling by amino acids in cell culture (SILAC) has allowed unambiguous identification of protein targets of lipid binding and modifications directly in native biological systems (34, 35). Given that baicalin does not contain any obvious reactive moiety that can modify proteins covalently, we designed and synthesized a photoaffinity baicalin probe that contains a benzophenone photo–cross-linking group and an alkynyl reporter group (Fig. 2A and SI Appendix, Fig. S3A). Both functional groups were introduced at the carboxylic acid group using a recently developed “photo-click” reaction (36). Despite considerable structural derivatizations, the baicalin probe was able to lower the lipid accumulation as effectively as the native compound (Fig. 2B). When the baicalin probe (200 μM) was used to label the proteome in an in-gel fluorescence assay, cotreatment of the native baicalin compound (400 μM) was able to compete the probe labeling (SI Appendix, Fig. S3B), supporting that the probe binds to similar target proteins as baicalin does.

To identify these target proteins, we performed a series of SILAC-ABPP experiments using the baicalin probe with or without competition from the native baicalin compound (Fig. 2C). Briefly, proteomes lysed from the “light” and “heavy” HeLa cells treated with excessive fatty acids were first incubated for 15 min with DMSO or baicalin (1 mM), respectively, and labeled with the baicalin probe (200 μM) via UV-induced photo–cross-linking. The probe-labeled light and heavy proteomes were mixed, conjugated with azide-biotin, enriched on streptavidin beads, and finally subjected to trypsin digestion. The digested peptides were analyzed by liquid chromatography tandem mass spectrometry to identify proteins that are photocaptured by the baicalin probe. The quantified SILAC ratio (light vs. heavy) for each protein reflects the extent of competition by the native baicalin compound, with a larger ratio corresponding to greater competition. In addition to the “Competition” experiment, we performed three sets of control experiments (benzophenone probe “BP-control,” control probe “CP-control,” and “UV-control”) to eliminate false-positive targets resulting from indirect and/or nonspecific binding to the probe (Fig. 2C and SI Appendix, SI Materials and Methods). Three replicates were performed for each of the control and competition experiments, and all of the proteins quantified are listed in Dataset S1. After applying a cutoff of 4.0 for the averaged SILAC ratio in all of the control and competition experiments, we collectively identified 142 proteins as potential targets that are specifically bound to baicalin (Fig. 2D and Dataset S2).

Gene ontology analysis (37) of the pathways associated with these 142 proteins revealed a top-ranking functional cluster of “fatty acid degradation,” which includes seven enzymes (Fig. 2E). Since this specific functional cluster of enzymes is highly relevant to the observed phenotype of reduced lipid accumulation, we continued to investigate these enzymes’ potential roles in mediating baicalin’s lipid-reducing effect. We used RNA interference (RNAi) to transiently knock down each of these enzymes (SI Appendix, Fig. S4) and observed that knockdown of CPT1A, the rate-limiting enzyme directly involved in FAO, completely abolished the effect of baicalin in reducing lipid accumulation (Fig. 2F). Consistently, cotreatment with malonyl-CoA, an endogenous CPT1A inhibitor, also significantly impaired the effect of baicalin (Fig. 2G). These results collectively suggested that the CPT1A-controlled FAO is responsible for the lipid-reducing effect of baicalin. Since abnormal lipid accumulation results from an imbalance of lipid metabolism and excessive lipid biogenesis and/or insufficient lipid consumption, we also performed the transcriptome analysis of the baicalin-treated cells and found no significant changes at the transcriptional level for those enzymes associated with various lipid metabolic pathways, including CPT1A (SI Appendix, Fig. S5 and Dataset S3). Thus, we hypothesized that baicalin might exert its lipid-reducing effect by directly activating CPT1A to accelerate lipid consumption in FAO.

To test whether baicalin can activate CPT1A directly, we first established a quantitative enzymatic assay to measure CPT1A activity, which uses the [D9]-carnitine and palmitoyl-CoA as the substrates and detects the production of [D9]-palmitoyl-carnitine (SI Appendix, SI Materials and Methods). In this assay, lysates from the baicalin-treated HeLa cells (100 μM, 24 h) showed a threefold increase in CPT1A activity compared with that of the DMSO-treated cells (Fig. 3A). Immunoblotting analysis confirmed that baicalin treatment did not perturb the protein level of CPT1A compared with the untreated cells (Fig. 3B). Direct addition of baicalin (100 μM) to the cell lysates resulted in an approximately sevenfold increase of CPT1A activity (Fig. 3C). Furthermore, treatment of the isolated mitochondria extracts with 100 μM baicalin resulted in a similar stimulation of CPT1A activity (Fig. 3D), which is consistent with the mitochondrial location of CPT1A. Similar CPT1A-activating effects can also be observed in primary hepatocytes (SI Appendix, Fig. S2B). We overexpressed CPT1A recombinantly in Escherichia coli (SI Appendix, Fig. S6) but could not obtain the active enzyme in the purified form due to protein aggregation. Nevertheless, we observed a greater than threefold increase in CPT1A activity from the E. coli lysates treated with 100 μM baicalin, whereas the background CPT1A activities were negligible (Fig. 3E). The activation of CPT1A by baicalin was dose-dependent, as we observed a >60-fold activation when the lysates were treated with up to 800 μM baicalin (SI Appendix, Fig. S7). We also recombinantly overexpressed CPT1A of Mus musculus and CPT1 of Caenorhabditis elegans and demonstrated that baicalin is able to stimulate these orthologs as well, although to a lesser extent (Fig. 3 F and G). Interestingly, baicalin did not activate CPT1B and CPT1C, the other two human isoforms of CPT1 (Fig. 3 H and I). These data collectively demonstrated that baicalin activates CPT1A directly with specific isoform selectivity.

Biophysical analysis of the interaction between CPT1A and baicalin can provide more insights on the mechanism of enzyme activation by the flavonoid. Since we could not obtain purified CPT1A with enzymatic activity, we resorted to other complementary methods to analyze the protein–flavonoid interaction directly in cell lysates. Both temperature- and dose-dependent cellular thermal shift assays (38) demonstrated that baicalin affects the thermal stability of CPT1A but not the actin control (Fig. 4 A and B and SI Appendix, Fig. S8), supporting that there is a direct interaction between CPT1A and baicalin. We next used the baicalin probe to map its potential binding site on CPT1A by a competitive ABPP reductive dimethylation experiment (39) (SI Appendix, SI Materials and Methods and Fig. S9A). Based on the CPT1A peptides whose probe labeling was sensitive to baicalin treatment or not (Fig. 4C and SI Appendix, Fig. S9B), we used the Rosetta software suite (40) to construct a model of the CPT1A–baicalin complex by homology modeling and ligand docking (Fig. 4D). Guided by the docking model, we designed four mutations in CPT1A at the complex interface (L286W, I291F, E309Y, and H327E) that are predicted to maintain the structural integrity of CPT1A but disrupt the binding of baicalin (SI Appendix, Fig. S10A). All of the four disruptive mutants were confirmed with much reduced labeling by the baicalin probe (Fig. 4E and SI Appendix, Fig. S10B). We tested their enzyme activity in response to baicalin treatment in vitro and found that all of them significantly lost the activation by baicalin (Fig. 4F). To further verify the role of CPT1 in mediating the lipid-reducing activity of baicalin, we overexpressed either wild-type CPT1A or each of the four mutants in HeLa cells after knocking down endogenous CPT1A expression by RNAi (SI Appendix, Fig. S11A) and evaluated the lipid-reducing activity of baicalin on them. As expected, only overexpression of wild-type CPT1A, but not any of the mutants, could rescue the phenotypic reduction of lipid accumulation in cells by baicalin (Fig. 4G), and the response is well correlated with the overall CPT1A enzymatic activity upon baicalin treatment (Fig. 4H).

We next investigated the effect of baicalin in a DIO animal model. Mice were fed a high-fat diet for 12 wk to first induce obesity and hepatic steatosis; Under the same high fat diet, they were then administered intragastrically with baicalin (400 mg/kg daily) for another 12 wk. The dose was chosen based on a previous pharmacokinetic study that measured the final plasma concentration of baicalin as 0.8 μg/mL 6 h after intake (41). As expected, DIO mice gained significantly more body weight (42 ± 4.7 g) than those on regular chow (30 ± 1.7 g) after 12 wk (Fig. 5A). Ensuing treatment with baicalin resulted in a dramatic reduction of whole-body weight (Fig. 5A), lowered the percentage of body fat (Fig. 5B), improved insulin sensitivity, and ameliorated hyperlipidemia and hyperglycemia (SI Appendix, Fig. S12). Without affecting the amount of food intake (Fig. 5C), baicalin treatment increased energy expenditure in DIO mice (Fig. 5D) and shifted the respiratory exchange ratio (RER), suggesting that these mice use more lipid as the energy source (Fig. 5E). Notably, the RER shift occurred as early as 3 wk after baicalin treatment and preceded the divergence of body weights (SI Appendix, Fig. S13), suggesting that activated FAO is driving the improvement in the metabolic profiles of DIO mice.

Analysis of the raw liver tissues showed that baicalin treatment effectively reduced the size and weight of livers (Fig. 5 F and G) as well as the accumulation of hepatic lipid (Fig. 5 H and I) in DIO mice. Furthermore, quantitative analysis of the tissue homogenates and blood samples showed that baicalin could significantly reduce the levels of hepatic triglyceride and cholesterol (Fig. 5J) and improve liver function (SI Appendix, Fig. S12B) in DIO mice. It should be noted that while baicalin could ameliorate hepatic steatosis in DIO mice, it showed no obvious adverse effects on the lean mice, such as weight reduction, loss of appetite, and hepatotoxicity (Fig. 5 A and C and SI Appendix, Fig. S12B), suggesting that it could be used safely to treat hepatic steatosis and obesity.

We measured the CPT1A activity from the homogenized liver lysates and found that baicalin treatment resulted in a greater than twofold increase in CPT1A activity (Fig. 5K) with little change on protein abundance or the level of malonyl-CoA (SI Appendix, Fig. S14). Consistent with activated FAO, the plasma level of nonesterified fatty acid (NEFA) was decreased and that of total ketone bodies (TKBs) was increased (Fig. 5L).

To corroborate the role of CPT1A activation in mediating the antiobesity effect of baicalin, we proceeded to test the baicalin-insensitive mutants in the DIO mouse model. We first examined the four baicalin-insensitive mutations in the mouse enzyme as they were designed based on the docking model from the human ortholog. The in vitro CPT1A assay revealed that H327E retained its resistance toward activation by baicalin (Fig. 6A). Consistently, sequence alignment of the human and mouse CPT1A showed that H327 is conserved among the four mutated sites (SI Appendix, Fig. S15). A thermal shift assay also confirmed that the H327E mutant of mouse CPT1A was no longer bound by baicalin (Fig. 6 B and C).

Since genetic knockout of CPT1A is embryonically lethal (42), we first knocked down the hepatic CPT1A expression in DIO mice using shRNA interference and observed that all beneficial effects of baicalin in ameliorating diet-induced hepatic steatosis and obesity were effectively abolished (SI Appendix, Fig. S16). To further demonstrate that activation of CPT1A is critical for these beneficial effects, we restored the level of CPT1A in liver by overexpressing either the wild-type CPT1A or the H327E mutant via adenoassociated virus-mediated transfection (SI Appendix, SI Materials and Methods). Consistent with our observation in the cellular model, only rescue with the wild-type CPT1A, but not the baicalin-insensitive mutant H327E, was able to ameliorate the obesity-associated metabolic symptoms in DIO mice, including reduced body weight (Fig. 6D) and percentage of body fat (Fig. 6E), reduced liver size, weight and lipid deposition (Fig. 6 F and G), decreased triglyceride and cholesterol levels in the liver (Fig. 6H), decreased NEFA and increased TKB levels in the plasma (Fig. 6I), and shifted metabolic rates (Fig. 6 J and K). In addition, baicalin ameliorates insulin sensitivity, hyperlipidemia, and hyperglycemia with dependence only on wild-type CPT1A but not the baicalin-insensitive mutant H327E (SI Appendix, Fig. S17). We confirmed that the “rescuing” effect was not due to different expression levels of the wild-type and mutant CPT1A (SI Appendix, Fig. S11B), but rather correlated with the overall enzymatic activity of CPT1A upon baicalin treatment (Fig. 6L). We also measured the abundance and activity of CPT1A in other tissues, such as adipose tissue, intestine, and kidney. Although CPT1A could be activated by baicalin across all these tissues, the overall activity of hepatic CPT1A was measured as the highest one (SI Appendix, Fig. S18). These data collectively supported that a direct interaction between baicalin and hepatic CPT1A is responsible for mediating the antiobesity and antisteatosis activity of the flavonoid.