Intermittent Fasting–Improved Glucose Homeostasis Is Not Entirely Dependent on Caloric Restriction in db/db Male Mice

According to the IDF Diabetes Atlas (10th edition), the global prevalence of diabetes reached 10.5% in 2021, and ∼6.7 million people died as a result of diabetes or its complications in 2021, accounting for ∼12.2% of all deaths worldwide (1). Obesity, an important risk factor for diabetes (2), is primarily caused by a long-term imbalance between energy intake and expenditure. Guidelines uniformly emphasize that dietary intervention, as a component of treatment, is a crucial aspect of diabetes management (3,4). However, there is currently no consensus on the dietary approach that should be adopted by patients with type 2 diabetes or on the optimal functional ratio of the three major dietary nutrients.

Studies have shown that caloric restriction (CR) improves the life span and metabolic health of several model organisms (3–6). However, it is difficult to implement prolonged CR in humans, because strict calorie counting is required, which can be challenging for most individuals and lead to psychological distress (6). Consequently, alternative feeding regimens, such as intermittent fasting (IF), have become increasingly popular (7). Clinical studies involving IF and CR have showed similar effectiveness for body weight, fat mass, fasting glucose, and insulin levels (8–12). Some studies have shown that IF is expected to replace CR in the treatment of type 2 diabetes because of the greater improvements seen in insulin sensitivity (13,14). The IF pattern involves prolonged fasting intervals with CR. Notably, IF can reduce total calorie intake by 30% (15), and the energy intake with IF and CR is reportedly consistent. Therefore, it remains unclear whether the effects of IF result from CR or from the long fasting period and alternating feeding diet pattern. It is critical to draw clear conclusions regarding these factors and clarify their underlying mechanisms.

However, in animal models, a typical CR model is inappropriate as a reference for distinguishing between CR and dietary patterns in IF. In most rodent CR studies, animals typically eat only once per day, rapidly consuming their food, which collaterally imposes a fasting period of ∼20–22 h after a meal is finished within 2–4 h (16). In addition, some overlap exists between the typical CR and IF patterns, narrowing the difference. Therefore, a CR murine model called meal feeding (MF) was established by feeding mice twice daily to reduce the fasting interval as a more appropriate reference for IF to distinguish CR from diet patterns. Every-other-day feeding (EODF) is the most commonly used research model for IF and has been applied in many clinical studies, even in individuals with diabetes, without serious adverse effects (17–21). Therefore, we used the MF model, where food intake was calculated based on the total consumption of the EODF group, distributed equally at 10:00 a.m. and 6:00 p.m., effectively reducing the fasting duration to ∼12 h. This strategy guaranteed precise management of the intake variable and minimized extended fasting periods, which is a novel aspect of our experimental design.

Here, we aimed to identify the specific elements of IF that mediate the aforementioned metabolic benefits. Our results might deepen our understanding of the mechanisms underlying IF, creating a foundation for future dietary interventions in clinical practice.

Eight-week-old male type 2 diabetic (BKS-Lepr^em2Cd479/Nju; db/db) and wild-type (WT; C57BLKS/JGpt; wt/wt) mice were purchased from GemPharmatech, Co., Ltd. Mice were acclimated to the animal facility conditions for 1 week. Diabetic mice were defined as having blood glucose >13.9 mmol/L for 2 consecutive days.

Initially, both db/db and wt/wt mice were divided into four groups (WT–ad libitum [WT-AL], WT-EODF, diabetic-AL [DB-AL], and DB-EODF; n = 8 each) for 28 days (Fig. 1A). Next, we focused on db/db mice, which were further subdivided into three groups (AL, MF, and EODF; n = 12 each) for 32 days (Fig. 2A). Additional mice were euthanized at the end of the first, second, and third weeks for every group (n = 6) to measure liver protein levels. In the third phase, db/db mice were injected with an adeno-associated virus 8 (AAV8) expressing Kruppel-like factor 9 (KLF9) under the control of the thyroxine binding globulin promoter (AAV8-Klf9) via the tail vein. Two weeks postinjection, the mice were divided into six groups (AAV-control-AL [AAV-CON-AL], AAV-CON-MF, AAV-CON-EODF, AAV-Klf9-AL, AAV-Klf9-MF, and AAV-Klf9-EODF; n = 6 each) and subjected to the respective dietary interventions for 32 days (Fig. 5A). The daily food intake per cage was measured to calculate the average daily food intake per mouse. Mice were weighed at 2-day intervals, and fasting blood glucose was measured at 6- or 8-day intervals using an ACCU-CHEK Performa glucometer (Roche). All animal experiments were approved by the Animal Research Committee of Sun Yat-sen University (approval no. SYSU-IACUC-2021-000228).

The mice were fed a standard chow diet obtained from the Guangdong Medical Laboratory Animal Center. The AL group had unrestricted access to food and water, whereas the EODF group fasted for 1 day, followed by AL feeding with free access to water (22,23). The MF group food intake was adjusted to achieve a similar CR, calculated relative to the total food intake of the EODF group. The daily food allocation to the MF group was evenly split at 10:00 a.m. and 6:00 p.m.

Before blood glucose level measurement, the mice were fasted for 8 h. On the 28th day, glucose tolerance testing was conducted after the mice had fasted for 8 h. On the 32nd day, an insulin tolerance test or pyruvate tolerance test was performed after the mice had fasted for 6 h. Glucose (1 g/kg), insulin (1 units/kg), or sodium pyruvate (1.25 g/kg) was administered via intraperitoneal injection. Glucose levels were measured at 0, 30, 60, 90, and 120 min after injection.

All mice were fed AL before euthanasia. After fasting for 4 h, the mice were humanely euthanized, and blood was sampled from the orbital vein. The tissue of the liver, epididymal white adipose tissue, and muscle tissue were collected. Serum cholesterol, triglyceride, insulin, glucocorticoid (GC), free triiodothyronine (FT3), and liver glycogen levels and glucose-6-phosphatase (G6pc) and phosphoenolpyruvate carboxykinase activity were measured according to the manufacturer’s instructions. All reagents and manufacturers are listed in Supplementary Table 1.

Liver samples were placed in 4% paraformaldehyde, dehydrated, and paraffin embedded. Subsequently, 4-μm-thick tissue slices were stained with hematoxylin-eosin and periodic acid Schiff. Frozen liver sections were stained with oil red O. Stained samples were imaged under an orthostatic optical microscope (Nikon NI-U).

RNAiso Plus (Takara Bio) was used to extract total RNA. cDNA was synthesized using 1,000-ng RNA samples with Evo M-MLV RT Premix (Accurate Biotechnology) according to the manufacturer’s protocol. The SYBR Green Pro Taq HS qPCR Kit (Accurate Biotechnology) was used for quantitative real-time PCR conducted on a LightCycler 480 II real-time PCR instrument (Roche). Gene expression changes were analyzed by the 2^−ΔΔCt method and normalized to Actb. The primer sequences used are listed in Supplementary Table 2.

The Total Protein Extraction Kit (Invent) containing a protease and phosphatase inhibitor cocktail was used to extract tissue protein. A cell lysis buffer (Cell Signaling Technology) was used for cell protein extraction. Proteins of ∼40 μg were separated in SDS-PAGE gels and transferred to polyvinylidene fluoride membranes (Millipore). Next, the membranes were blocked with 5% nonfat milk (Sangon Biotech) in Tris-buffered saline with Tween at 25°C for 1 h and subsequently incubated with primary antibodies (diluted in Tris-buffered saline with Tween) at 4°C overnight and secondary antibodies at 25°C for 1 h. Protein bands were detected using the BeyoECL Star Kit (Beyotime Biotechnology). β-actin served for normalization. The primary antibodies involved are listed in Supplementary Table 3.

Tissue protein extraction was carried out as previously described. Protein A/G magnetic beads (MedChemExpress) at a volume of 100 μL were incubated with 2 μg GC receptor (GR) antibody (Proteintech) at 25°C for 2 h. The tissue lysate was then added to the beads-antibody complex and left to incubate overnight at 4°C. Once the mixture was washed with immunoprecipitation buffer, the beads were resuspended with loading buffer (1×) and heated at 95°C for 10 min. Finally, Western blotting was carried out as previously described, with ubiquitin antibodies as the primary antibody.

RNA extraction, transcriptome sequencing, and analysis were outsourced to OE Biotech Co., Ltd. Differential expression analysis was performed using DESeq2 to identify significantly differentially expressed genes (DEGs), setting a threshold of Q < 0.05, fold change >1.5, or fold change <0.67. A radar map of the top 20 genes was drawn to show the expression of DEGs using the R packet ggradar.

AML12 cells were cultured in DMEM/F12 medium (Gibco) supplemented with 10% FBS, 1× insulin-transferrin-selenium, and 100 nmol/L dexamethasone (24). The cells were maintained at 37°C in a 5% CO₂ cell culture incubator.

Lentiviral (LV) particles LV Sh-Control and LV Sh-Klf9 were purchased from GeneChem. AML12 cells were infected according to the manufacturer’s instructions and incubated in complete medium for 48 h. For stable selection, the infections were treated with puromycin dihydrochloride selection marker. AML12 cells stably infected with LV Sh-Control or LV Sh-Klf9 were then used in the following experiments.

AML12 cells infected with LV Sh-Control or LV Sh-Klf9 were incubated with 250 μmol/L sodium palmitate and 33.3 mmol/L glucose for 24 h, followed by glucose-free medium for 1 h, and then supplemented with 10 mmol/L lactate and 1 mmol/L pyruvate for 4 h. Glucose in the medium was measured using the Glucose GOD-POD Kit (Elabscience) and normalized by protein levels. Concurrently, the cells were also collected for gene expression analysis.

Data are presented as mean ± SEM. Analyses were performed using Prism 8 (GraphPad Software). Statistical analyses were performed using one-way ANOVA, followed by Tukey-Kramer post hoc tests, and P values <0.05 were considered statistically significant.

The db/db and WT mice were divided into AL and EODF groups (Fig. 1A). After 28 days of intervention, the EODF group exhibited significant weight loss compared with the AL group in db/db mice, but not in WT mice (Fig. 1B). In both db/db and WT mice, the EODF group showed a cumulative food intake reduction of ∼35% compared with the AL group (Fig. 1C). Hyperglycemia was significantly alleviated after a 3-week intervention with EODF in db/db mice, but not in WT mice (Fig. 1D). No significant change was observed in the fasting insulin level in db/db or WT mice (Fig. 1E). Additionally, EODF significantly improved insulin resistance (Fig. 1E) and glucose tolerance in db/db mice, but not in WT mice (Fig. 1F and G). Overall, EODF improved glucose homeostasis in db/db mice, but not in WT mice. Considering that cumulative food intake was reduced by 35% in the EODF group compared with in the AL group, it remains unclear whether the metabolic benefit derived from CR or the unique diet patterns.

Another experiment was conducted in db/db mice to determine whether the metabolic benefits of EODF depended solely on CR (Fig. 2A). Although EODF mice ate 35% less than AL mice and matched the MF group intake, they lost less weight than MF mice but had greater reductions in fasting blood glucose (Fig. 2B–D). The EODF group demonstrated a more pronounced improvement in HOMA for insulin resistance compared with the AL group. However, there was a slight, nonsignificant improvement in HOMA-β within the EODF group (Fig. 2E). Furthermore, EODF showed more significant improvements in glucose tolerance and reduced glucose production from pyruvate than MF, with similar improvements in insulin sensitivity (Fig. 2F–H). These results, mirrored in lean mice, indicate that the benefits of EODF extend beyond mere calorie reduction, potentially offering superior control over gluconeogenesis (Supplementary Fig. 1).

Considering that db/db mice exhibit abnormal hepatic glucose output because of excessive glycogen use and gluconeogenesis activation compared with WT mice, we then focused on whether EODF affects liver glucose output in db/db mice. Compared with AL and MF groups, the EODF group demonstrated enhanced hepatic glycogen accumulation, evidenced by periodic acid Schiff staining and liver homogenate ELISA (Fig. 3A and B), and upregulated glycogen synthesis enzyme GYS2, with a slight nonsignificant downregulation of glycogen breakdown enzyme PYGL, according to Western blot and mRNA analyses (Fig. 3C and D). Moreover, EODF suppressed gluconeogenesis, reducing the activity of key enzymes PEPCK and G6pc (Fig. 3E) and their mRNA expression (Fig. 3F), suggesting the effectiveness of EODF in moderating hepatic glucose output independently of CR. Both EODF and MF improved liver insulin signaling, observed in enhanced phosphorylated AKT (p-AKT)/AKT and p-FOXO1 (Ser256)/FOXO1 ratios postinsulin (2 units/kg) injection (Fig. 3G), aligning with improved insulin sensitivity in insulin tolerance test results (Fig. 2G) (25,26). Additionally, we observed a similar increase in the p-AKT (Ser 473)/AKT ratio in skeletal muscle and adipose tissue for both MF and EODF (Fig. 3H). Furthermore, MF and EODF similarly reduced liver steatosis, lipid deposition (Supplementary Fig. 2A), and triglyceride levels (Supplementary Fig. 2B) without significantly affecting serum cholesterol or triglycerides (Supplementary Fig. 2D and E), implying that the liver benefits of EODF extend beyond insulin signaling and lipid metabolism improvement.

To better understand the mechanism by which EODF improves hepatic glucose output, liver mRNA sequences were constructed and examined. A total of 519 DEGs were found between EODF and AL groups and 231 DEGs between EODF and MF groups. By intersecting these DEGs, we identified 155 genes uniquely altered by EODF (151 upregulated and four downregulated), as shown in Fig. 4A. The Klf9 gene was selected for further analysis based on prior research.

KLF9 is a transcription factor known to promote hepatic gluconeogenesis by enhancing PGC1-α expression (27). In the EODF group, hepatic KLF9 and PGC1-α expression were significantly downregulated compared with in the MF and AL groups, with no marked difference between MF and AL, indicating that EODF-mediated downregulation of KLF9 may inhibit gluconeogenesis independently of CR (Fig. 4B and C). Using the AML12 cell line with shKLF9 transfection under high-glucose and palmitic acid conditions to mimic a diabetic state, we found that although these conditions upregulated Klf9, Pck1, and G6pc, Klf9 knockdown significantly reduced their expression and gluconeogenesis activity (Fig. 4D–F). Therefore, EODF may suppress gluconeogenesis through KLF9 and PGC1-α downregulation in the liver.

In our study, db/db mice were transfected with AAV8-Klf9 under the thyroxine binding globulin promoter to overexpress Klf9 in the liver and subjected to dietary interventions for 32 days (Fig. 5A). Despite similar food intake and body weight across dietary patterns, EODF reduced cumulative food intake by 40% compared with AL. Although MF led to slightly greater weight loss than EODF, both diets resulted in equal weight changes in AAV-Klf9– and AAV-CON–injected mice (Fig. 5B and C). EODF improved fasting blood glucose and glucose tolerance, suppressing gluconeogenesis to a greater extent than MF in AAV-CON–treated db/db mice, an effect not observed in AAV-Klf9 mice, indicating that Klf9 overexpression in the liver might negate the additional glucose metabolic advantages of EODF (Fig. 5D–I). Western blot analysis confirmed that KLF9 and PGC1-α were downregulated in the EODF group compared with in the MF and AL groups in AAV-CON db/db mice but not in AAV-Klf9 db/db mice (Fig. 5J). These results suggest that Klf9 overexpression in the liver limited the extra glucose metabolic benefits in EODF db/db mice.

KLF9 is regulated by the GC receptor (GR) in the liver, usually as a response to GC and FT3. Here, the protein levels of GR were downregulated in the livers of EODF mice compared with in MF and AL mice, whereas GR mRNA levels were similar across groups (Fig. 6A and B). Surprisingly, serum GC levels significantly increased (Fig. 6C), whereas FT3 levels remained unchanged, in the EODF group (Supplementary Fig. 2F). GCs are known to increase hepatic glucose output; however, gluconeogenesis was suppressed in EODF-treated mice, leading to conflicting findings. To confirm these results, the serum GCs and liver GR levels were measured at different time points. Sequential measurements showed serum GCs increased after 3 weeks of EODF, whereas liver GR levels dropped at 2 weeks (Fig. 6D and E), suggesting that the elevation in serum GC levels might be a compensatory response to reduced liver GR resulting from feedback regulation. Additionally, the decreased liver GR correlated with enhanced ubiquitination (Fig. 6F), whereas GR levels were not downregulated in other metabolic tissues, such as epididymal white adipose tissue and muscle (Fig. 6G).

Therefore, EODF seems to suppress gluconeogenesis by reducing hepatic GR via promotion of ubiquitination, which improves glucose homeostasis independently of CR (Fig. 7).

Our study suggests that EODF provides significant metabolic improvements in db/db mice, such as better glucose tolerance, reduced hyperglycemia, and lower hepatic glucose output, benefits that extend beyond mere CR and may be due to the particular diet pattern. Additionally, EODF specifically triggered liver GR ubiquitination and subsequent degradation, thereby possibly decreasing KLF9 and potentially contributing to gluconeogenesis suppression, ultimately leading to elevated serum GC levels from compensatory feedback mechanisms.

Several previous studies have reported similar phenotypes in glucose metabolic effects, including lower fasting glucose, comparing IF with CR diets (28–31). However, only a few studies have directly compared the metabolic benefits of these interventions. The typical CR model involves feeding once a day, which might prolong fasting intervals and was inappropriate for IF to distinguish the effects of calorie intake. Interestingly, Pak et al. (4) reported that prolonged fasting intervals were necessary for the metabolic benefits of a CR diet model, and Acosta-Rodríguez et al. (3) reported that in a CR model, the daily fasting interval and feeding circadian system functioned synergistically to prolong longevity in mice, underscoring the significance of fasting intervals in diet regimens. Taking this into account, we designed an MF model, matching the EODF group intake and dividing food intake into two sessions at 10:00 a.m. and 6:00 p.m. daily, which shortened the fasting interval to appropriately distinguish the calorie effects of EODF. This novel approach aimed to identify the specific elements of IF that yield benefits and potentially inform future dietary interventions in clinical settings.

Nonoverweight individuals are increasingly adopting IF diets in their daily lives. This study showed that EODF can improve glucose regulation, resulting in significant weight loss in db/db mice without significantly affecting normal lean mice. Previous studies have demonstrated that EODF can lead to weight loss of 0.2–0.5 kg per week in obese and overweight individuals (BMI ≥25 kg/m²) (9,12,32,33). However, for those with normal body weight (18.5 kg/m² ≤ BMI < 25 kg/m²), the weekly weight loss was only 0.2 kg, without statistical significance (34). This finding parallels our observations, underscoring the potential of EODF as an effective intervention for obesity and diabetes, although its efficacy in lean mice is comparatively limited. Intriguingly, the EODF group did not exhibit as much weight loss as the MF group. However, the glucose homeostasis enhancement observed in db/db mice was not exclusively attributed to CR, indicating that the influence of EODF extends beyond CR.

Further research is necessary to elucidate how EODF improves glucose metabolism and yields the advantages of IF beyond mere calorie reduction. Investigations involving liver mRNA sequencing and Klf9 overexpression in db/db mice suggest that KLF9 is integral to the additional metabolic benefits of EODF. Whereas GCs reportedly induce Klf9 expression, enhancing liver gluconeogenesis and hyperglycemia, our findings indicate a reduced Klf9 expression in the livers of EODF-subjected db/db mice (27). GCs mediate their physiological effects through binding to GRs, which then translocate to the nucleus to bind to GC response elements on the promoters of Klf9 (35,36), which in turn activate Pgc1α gene expression and the gluconeogenic program (27,37). In our study, the multicycle fasting-refeeding dietary pattern promoted ubiquitination and degradation of liver GR while preventing GC-induced Klf9 transcription and gluconeogenesis. In addition, db/db and ob/ob mice showed higher liver KLF9 levels compared with normal C57BL/6J mice (27). So far, there is no literature confirming the changes of KLF9 in db/db or ob/ob mice after fasting. Therefore, db/db mice may have different KLF9 changes after fasting compared with normal C57 mice because of differences in basal KLF9 levels and pathological status of glucose intolerance. Additionally, hyperglycemia-induced KLF9 represses the antioxidant PRDX6, leading to oxidative stress and mitochondrial disorders in the neuron system (37). KLF9 has also been found to aggravate cardiomyopathy in diabetes by inhibiting PPARγ/NRF2 signaling (38). Furthermore, Klf9 knockdown reportedly relieved gestational diabetes by upregulating DDAH2 and alleviating inflammation and oxidative stress (39). These insights reinforce the critical role of KLF9 in mediating the beneficial effects of IF beyond mere caloric intake reduction.

Notably, in our study, compared with the MF group, the EODF group was solely controlled for caloric intake, but there were differences in fasting duration and circadian rhythms, which may be the alternative mechanisms explaining the benefits of EODF. Throughout history, humans have experienced periods of starvation, leading to an evolutionary adaptation to long-term fasting. Fasting has been shown to offer metabolic advantages, including increasing liver FGF21 expression (40,41), promoting autophagy, helping remove dysfunctional organelles and metabolic waste, and maintaining physiological cell function (42–44). Furthermore, evidence suggests that EODF may influence the circadian rhythms of peripheral organs. Notably, peripheral organs such as the liver possess intrinsic biological clocks that are regulated by both the central circadian clock and external factors, including dietary patterns and sleep cycles (45–47). This regulatory mechanism implies that fasting interventions, like EODF, could modulate these peripheral rhythms. Supporting this, research on time-restricted feeding, which is a variant of IF, has demonstrated that such dietary strategies can enhance metabolic balance in mice with altered circadian clocks (44,48). These findings provide a plausible mechanism by which EODF might adjust the metabolic rhythms of peripheral organs, offering insights into the beneficial effects observed in our study.

The use of EODF may have adverse effects, limiting its metabolic benefits. GC is a common hormone with a variety of physiological functions, including immunity regulation, promotion of liver gluconeogenesis, protein decomposition, and electrolyte homeostasis maintenance. However, high levels of GC can promote insulin resistance and even accelerate the development of diabetes (47–52). Our data suggest that the liver counteracts the hyperglycemia-promoting effects of GCs by reducing GR levels, but this effect was not observed in other metabolic tissues, particularly muscles. Therefore, we propose that under EODF conditions, muscle and adipose tissues may respond to elevated GC levels, potentially leading to adverse effects such as muscle loss and limiting glucose metabolism benefits. Clinical studies support this, showing greater muscle loss with EODF compared with MF in individuals with BMI <24 kg/m² (52). Our study presents an intriguing correlation between hepatic GR levels and serum GC levels in response to EODF. The temporal sequence of GR decrease and GC increase suggests a potential causal relationship, with hepatic GR downregulation possibly preceding and contributing to elevated serum GC levels. However, this remains speculative, and additional investigations are necessary to confirm this causal relationship and fully unravel the underlying effects.

Nonetheless, it is important to acknowledge certain limitations. Firstly, the translation of EODF from mice to humans is challenging because of differences in life span and metabolism. For instance, a 24-h fasting period in mice may be equivalent to a 5-day fasting period in humans (47). Moreover, ketone body regulation during fasting varies between mice and humans (51,53). Additionally, our findings did not definitively establish a connection between reduced GR protein levels and the observed metabolic benefits or decreases in KLF9 under EODF. The challenge of restoring GR levels in EODF mice to match those in the AL and MF groups, particularly considering the intricacies of ubiquitin-mediated degradation, is an area for future research. Furthermore, we shall investigate the effects of different fasting intervals, the long-term effects of IF and CR, and the impact of various diets on glucose homeostasis and metabolic outcomes. However, our experimental conditions did not allow for more extensive methods, such as glucose clamp experiments and isotope tracing, to enhance our conclusions regarding gluconeogenesis.

In conclusion, our research underscores the effectiveness of EODF in improving glucose homeostasis in db/db mice, with benefits that surpass those of simple CR and are attributed to a specific dietary pattern. EODF uniquely modulates the ubiquitination of hepatic GR, thereby reducing KLF9 levels and ultimately suppressing gluconeogenesis. Clinically, these findings emphasize the potential of EODF as an innovative dietary approach to manage and prevent glucose-related metabolic disorders.

Acknowledgments. The authors thank Professor Phei Er Saw for writing guidance the three anonymous reviewers for their insightful comments and constructive suggestions on the prior versions of this manuscript, everyone who provided help in the Department of Endocrinology and the authors’ laboratory, and eBioart for help in pattern diagram making.

Funding. This study was supported by the National Natural Science Foundation of China (U20A20352), Guangdong Clinical Research Center for Metabolic Diseases (2020B1111170009), Guangzhou Key Laboratory for Metabolic Diseases (202102100004), Guangdong Basic and Applied Basic Research Foundation (2019A1515011199, 2021B1515020005, and 2023A1515030079), Guangzhou Science and Technology Program Key Projects (202206010031), and Guangdong Science and Technology Department (2020B1212060018 and 2020B1212030004).

The funders were not involved in the design or analysis of this research, the preparation of the manuscript, or the decision to publish.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.Z., X.Ho., and X.He performed most of the experiments and analyses and wrote the manuscript. J.L., S.F., J.W., and Z.L. performed some experiments. S.C. contributed to data interpretation and discussion. L.Y., M.R., and W.W. contributed to the conception and design of study. All authors reviewed and approved the final version of the manuscript and gave approval for this version to be published. W.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.