Metabolic Consequences of Neuronal HIF1α-Deficiency in Mediobasal Hypothalamus in Mice

Objective This study aims to investigate whether hypoxia-inducible factor 1α (HIF1α) in the neurons of the mediobasal hypothalamus is involved in the regulation of body weight, glucose, and lipid metabolism in mice and to explore the underlying molecular mechanisms. Methods HIF1α flox/flox mice were used. The adeno-associated virus that contained either cre, GFP and syn, or GFP and syn (controls) was injected into the mediobasal hypothalamus to selectively knock out HIF1α in the neurons of the mediobasal hypothalamus. The body weight and food intake were weighed daily. The levels of blood glucose, insulin, total cholesterol (TC), triglyceride (TG), free fatty acid (FFA), high-density lipoprotein (HDL), and low-density lipoprotein (LDL)were tested. Intraperitoneal glucose tolerance test (IPGTT) was performed. The insulin-stimulated Akt phosphorylation in the liver, epididymal fat, and skeletal muscle were examined. Also, the mRNA expression levels of HIF1α, proopiomelanocortin (POMC), neuropeptide Y (NPY), and glucose transporter protein 4 (Glut4) in the hypothalamus were checked. Results After selectively knocking out HIF1α in the neurons of the mediobasal hypothalamus (HIF1αKOMBH), the body weights and food intake of mice increased significantly compared with the control mice (p < 0.001 at 4 weeks). Compared with that of the control group, the insulin level of HIF1αKOMBH mice was 3.5 times higher (p < 0.01). The results of the IPGTT showed that the blood glucose level of the HIF1αKOMBH group at 20–120 min was significantly higher than that of the control group (p < 0.05). The serum TC, FFA, HDL, and LDL content of the HIF1αKOMBH group was significantly higher than those of the control group (p < 0.05). Western blot results showed that compared with those in the control group, insulin-induced AKT phosphorylation levels in liver, epididymal fat, and skeletal muscle in the HIF1αKOMBH group were not as significantly elevated as in the control group. Reverse transcription-polymerase chain reaction (RT-PCR) results in the whole hypothalamus showed a significant decrease in Glut4 mRNA expression. And the mRNA expression levels of HIF1α, POMC, and NPY of the HIF1αKOMBH group decreased significantly in ventral hypothalamus. Conclusions The hypothalamic neuronal HIF1α plays an important role in the regulation of body weight balance in mice under normoxic condition. In the absence of hypothalamic neuronal HIF1α, the mice gained weight with increased appetite, accompanied with abnormal glucose and lipid metabolism. POMC and Glut4 may be responsible for this effect of HIF1α.

Conclusions: The hypothalamic neuronal HIF1a plays an important role in the regulation of body weight balance in mice under normoxic condition. In the absence of hypothalamic neuronal HIF1a, the mice gained weight with increased appetite, accompanied with abnormal glucose and lipid metabolism. POMC and Glut4 may be responsible for this effect of HIF1a.
Keywords: HIF1a in the hypothalamus, obesity, glucose, lipid metabolism, insulin resistance BACKGROUND Obesity refers to a metabolic disease in which the energy intake exceeds the energy expenditure, leading to increased body weight and body fat content. In 2015, 107.7 million children and 603.7 million adults were obese in the world, and about 1.9 billion were overweight (1,2). The epidemics of obesity and obesity-related diseases, such as dyslipidemia, hypertension, and type 2 diabetes, are rapidly increasing worldwide (3,4). Researchers never stopped exploring the treatment of obesity, but to date, no absolute safe and effective weight loss method or drug have been found. Therefore, studying the pathogenesis of obesity from different perspectives and finding possible treatments are still necessary.
The hypoxia-inducible factor (HIF) is a transcription factor that is stably expressed in hypoxic cells and participates in various adaptive responses (5,6). Three isoforms of HIF, namely, HIF1a, HIF2a, and HIF3a, are identified. Studies showed that HIF1a and HIF2a are stably expressed in the brain, including hypothalamus under normoxic condition (7,8). Moreover, studies showed that the hypothalamic HIFs play an important role in appetite regulation in mice. Studies demonstrated that increased hypothalamic glucose level activates HIF and inhibits feeding in mice (7). The knockdown of both HIF1a and HIF2a in proopiomelanocortin (POMC) neurons or arcuate nucleus (ARC) results in pronounced weight gain under high-fat diet (HFD) (7,9). The HIF2a expression decreases with the aging of the mice, and the deletion of HIF2a in hypothalamic POMC neurons can lead to age-dependent weight gain and increased body fat content in mice with mild glucose intolerance and insulin resistance (8). These studies suggested that HIFs, especially HIF2a, in hypothalamic ARC, precisely POMC neurons may be involved in appetite regulation and body weight regulation in mice fed with HFD. However, it is not clear whether hypothalamic neuronal HIF1a itself plays a role in body weight regulation in mice, especially under normal diet and whether it affects the glucose and lipid metabolism.
Therefore, in our study, HIF1a flox/flox mice were selected, and adeno-associated viruses containing cre fragments and neuronspecific promoter syn were injected into mediobasal hypothalamus to knock out HIF1a selectively in the neurons of the mediobasal hypothalamus. The body weight and food intake of the mice are recorded under chow diet. Also, the glucose metabolism and the lipid metabolism of the mice were examined. As a first step to understand the role of HIF1a in body weight control, we aim to identify the metabolic phenotype of neuronal HIF1a-deficiency in mediobasal hypothalamus in mice.

Animals
Breeding pairs of HIF1a flox/flox mice were gifts from Polotsky Laboratory (10) of Johns Hopkins University and bred at the Animal Experimentation Center of Xinjiang Medical University. Male HIF1a flox/flox mice aged 6-8 weeks and weighing approximately 23 g were used for the study. Male C57BL/6J mice aged 6-8 weeks were used to test the effect of viruses on body weight. All mice had free access to water and were housed in a standard specific pathogen free (SPF)-grade laboratory environment at 22°C-23°C with a 12-h light/dark cycle (09:00-21:00/21:00-09:00). Mice were anesthetized by intraperitoneal injection of 4% chloral hydrate (0.13 ml/10 g) for all surgeries. After virus injection, up to five mice were kept in one cage on chow diet (9.4% kcal from fat) (11), and cages were changed twice a week. The daily food intake of mice is calculated by cage: the total food intake of the cage/the number of mice. All animal experiments were approved by the Animal Ethics Committee of Xinjiang Medical University and conducted in accordance with the guidelines established by this committee.

Experimental Design
The mice with significant weight gain were divided into three batches 28-35 days after the virus injection. In the first batch, the brain of the mice was harvested after whole body perfusion for frozen section. Intraperitoneal glucose tolerance test (IPGTT) and insulin signaling pathway were examined in the mice of the second batch. The blood and the fresh hypothalamus were harvested in the mice of the third batch for future use.

Frozen Sections to Observe the Location of Injections and Expression of Viruses
The HIF1a flox/flox mice were anesthetized and rapidly perfused with sterile normal saline and 4% paraformaldehyde 28-35 days after virus injection. The brains were carefully removed, postfixed in 4% paraformaldehyde at 4°C for overnight. The next morning, the brains were cytoprotected in 20% sucrose solution at 4°C for 24 h. Then, the brains were frozen and stored at −80°C. Whole brains were fixed with the O.C.T. compound and 30 µm of sections were cut and stored in PBS buffer. Then the brain slices were mounted on the slide. After sealing with antifade solution (Solarbio, S2100, Beijing, China), the injection location and virus expression were observed and pictured by confocal microscope (Leica, SP8, Wetzlar, Germany) with fullslice scanning using ×10 magnification.

IPGTT
IPGTT were performed after 6 h fasting. The glucose solution (1 g/kg) (BIOFROXX, 1179GR500, Einhausen, Germany) was intraperitoneally injected into the mice after testing blood glucose at baseline. The blood glucose was measured at the tip of the tail using a handheld glycemic meter (Roche Accu-Chek, Active, Indianapolis, IN, USA) at 10, 20, 30, 60, 90, and 120 min after injection of glucose.

Biochemical Measurements
The triglyceride (TG) assay kit (Elabscience, E-BC-K238, Wuhan, China) was used to detect TG content in serum. The total cholesterol (TC) ELISA kit (Elabscience, E-BC-K179) was used to detect the content of TC in serum. The mouse free fatty acids (FFA) ELISA kit (TW-reagent) was used to detect the content of FFA in serum. The glucose content in serum was detected using the mouse glucose ELISA Kit (Elabscience, E-BC-K268). The serum insulin content was measured using the mouse insulin ELISA Kit (Raybio, ELM-Insulin, Peachtree Corners, GA, USA).

Insulin Signaling Pathway Detection
Insulin signaling pathway activation in mice was performed by administering 5 U/kg insulin intraperitoneally 15 min prior to the sacrifice of the mice. In control mice, saline was injected. Liver, epididymal fat, and skeletal muscle tissue were collected, snap frozen in liquid nitrogen, and stored at −80°C. Tissues were homogenized through the Ripa buffer (ThermoFisher, VH310061, Waltham, MA, USA). The gels for Western blot were made by a gel-making kit (Solarbio, PC0020). Proteins (30 mg) were applied to each lane. For total and phosphorylated Akt measurements, we used antibody phospho-Akt (Ser473) (D9E) XP rabbit mAb (Cell Signaling, #4060, Danvers, MA, USA) and Akt (pan) (11E7) rabbit mAb (Cell Signaling, #4685). The optical densities of bands were measured using the ImageJ software. Insulin signaling was assessed by calculation of the ratio of pAkt to actin or total Akt after insulin injection.

Statistical Analyses
Mean ± SEM was used to calculate the results, and two-sided unpaired t-tests were used for both data sets. Pearson correlation tests for correlation between variables and normal distributions were performed using IBM SPSS Statistics 26 analysis software, as well as repeated measures ANOVA for weight and food intake for HIF1aKOMBH, C57BL/6J, and controls, and Bonferroni was used for multiple comparisons. GraphPad Prism 9 was used to calculate the area under the curve. Data are expressed as mean ± SEM, and p < 0.05 was considered statistically significant.

Increased Body Weight and Food Intake in HIF1aKOMBH Mice
No significant difference in mean body weight was found between the HIF1aKOMBH and control groups before virus injection. On 16 to 28 days after virus injection, the body weights of HIF1aKOMBH mice increased significantly compared with the control mice ( Figure 2B). The repeated measures ANOVA analysis showed that the interaction term was statistically significant, F (23, 897) = 17.880, p < 0.05, bias h 2 = 0.314, suggesting an interaction effect of group and time change on body weight in mice. By comparing the difference in body weight of mice seen in the two groups, the results showed that F = 7.006, p < 0.05, deviation h 2 = 0.152, indicating that the difference in body weight of mice in the two groups was significant. Twentyeight days after virus injection, the mean body weight of mice in the HIF1aKOMBH group reached 33.68 ± 1.43 g (n = 26), which was 1.3 times that of the control mice (26.01 ± 0.48 g, n = 15; Figure 2B). The maximum body weight of mice in the HIF1aKOMBH group reached 46.2 g (Figure 2A). For wildtype C57BL/6J mice, significant increases in body weight on days 24 and 27 after virus injection were seen in the cre group compared with the control group (p < 0.05) (n = 9, Figure 2D). However, the maximum body weight of the C57BL/6J mice in the cre group on day 27 after virus injection was only 30.9 g. The two-way ANOVA analysis showed that the interaction term for wild-type C57BL/6J mice in this study was statistically significant, F (8, 128) = 79.076, p < 0.05, deviation h 2 = 0.832, but there was no significant difference in weight between the two groups (F = 0.825, p > 0.05, deviation h 2 = 0.049).
The food intake of HIF1a flox/flox mice injected with AAVhsyn-cre-GFP significantly increased compared with that of the control mice. The two-way ANOVA analysis showed that the interaction term was statistically significant, F (23, 575) = 2.389, p < 0.05, bias h 2 = 0.087, suggesting an interaction effect of group and time change on the change of food intake. Statistical significance was observed between the two groups of mice from days 10 to 13, 15, and 17-29 (p < 0.05, p < 0.01, and p < 0.001, respectively; Figure 2C). The results of the comparison of the difference in the amount of food eaten by the mice between the different groups showed a significant difference in the weight of food eaten by the mice between the two groups (F = 10.914, p < 0.05, deviation h 2 = 0.304). For WT C57BL/6J mice, the number of days and group interaction had a significant effect on mouse feeding (F (5, 75) = 7.476, p < 0.05, deviation h 2 = 0.333). However, there was no difference in mouse feeding between groups (F = 0.044, p > 0.05, deviation h 2 = 0.003) (n = 9, Figure 2E).

Blood Glucose and Insulin Levels in HIF1aKOMBH Mice
Thirty days after AAV-hsyn-GFP and AAV-hsyn-cre-GFP injection in the mediobasal hypothalamus of HIF1a flox/flox mice, serum insulin and fasting blood glucose were detected. No significant differences were shown in fasting blood glucose level between HIF1aKOMBH mice (n = 8) and control mice (n = 7) ( Figure 3A). However, the serum insulin level of HIF1aKOMBH mice reached 49.70 ± 14.18 UIU/ml (n = 12), which was 4.9 times higher than that of the control mice (n = 23,  p < 0.01, Figure 3B). The correlation analysis in HIF1aKOMBH mice showed that the serum insulin level was positively correlated with weight gain (p < 0.05, r = 0.669, n = 11; Figure 3C).

Impaired Glucose Tolerance in HIF1aKOMBH Mice
IPGTT was performed 28-35 days after virus injection to investigate the glucose tolerance of mice. The results showed that both groups had peak blood glucose levels 20 min after glucose injection. There were no significant differences for the baseline glucose level of the two groups, similar with the fasting glucose level we tested before. However, HIF1aKOMBH mice had higher blood glucose levels at each time point after glucose injection than control mice, and the statistical significances were seen at 20, 30, 60, 90, and 120 min after glucose injection (n = 11; p < 0.05, p < 0.01, and p < 0.001, Figure 4A). The area under the curve (AUC) of the HIF1aKOMBH knockout mice (1,880.38 ± 138.69) was 1.2 times higher than that of the control mice (1,350.38 ± 49.51, p < 0.01; Figure 4B).

Attenuated Insulin Signaling Pathway in HIF1aKOMBH Mice
In control mice, p-Akt protein expression significantly increased in the liver, epididymal fat, and skeletal muscle in insulin injection group (n = 5; Figure 5), which implied a normal insulin signaling pathway in these mice. However, there was no significant difference in Akt phosphorylation in liver, epididymal fat, and skeletal muscle of HIF1aKOMBH mice between insulin injection group and saline injection group

Abnormal Lipid Metabolism in HIF1aKOMBH Mice
The investigation of TC, TG, FFA, high-density lipoprotein (HDL), and low-density lipoprotein (LDL) levels in the serum of HIF1aKOMBH mice showed that lipid metabolism was also abnormal in HIF1aKOMBH mice. No significant difference was found in the serum TG content between the two groups ( Figure 6A). The serum total cholesterol and FFA level in HIF1aKOMBH mice was significantly higher than control mice (p < 0.01 and p < 0.05, respectively, Figures 6B, C). For lipoprotein levels, the HDL and LDL content in HIF1aKOMBH mice was also significantly higher than those in control mice (p < 0.01 and p < 0.05, respectively, Figures 6D, E).

Changes in the Expression of Protein and Appetite-Related Neuropeptides Associated With Systemic Energy Homeostasis in HIF1aKOMBH Mice
Fresh hypothalamus were harvested 28-35 days after virus injections to detect the mRNA expression of HIF1a, POMC, NPY, and Glut4 by RT-PCR. Tests on the whole hypothalamus showed no significant changes in HIF1a (n = 16), POMC (n = 17), and NPY (n = 17) between two groups. However, a 2.05 times increase in Glut4 mRNA levels compared with the controls were seen (n = 18) ( Figure 7A). While tests on the ventral hypothalamus showed that the expression of HIF1a in HIF1aKOMBH mice (n = 4) was 74% less than the control group (n = 3) (p < 0.05), indicating that the neuronal HIF1a knockout in mediobasal hypothalamus of mice was successful. In ventral hypothalamus, the expression of POMC mRNA and NPY mRNA in HIF1aKOMBH mice (n = 4) was significantly reduced (p < 0.05) compared with the control group (n = 3), but the Glut4 expression in the hypothalamus of HIF1aKOMBH mice (n = 4) was not significantly different from the control group (n = 3) in ventral hypothalamus ( Figure 7B).

DISCUSSION
In the central nervous system, the hypothalamus is the basic center for the regulation of energy metabolism and has a powerful regulatory role in energy intake, energy expenditure, and weight control (12)(13)(14)(15), but the underlying mechanisms are unclear. Our study showed that HIF1a in neurons of mediobasal hypothalamus plays an important role in the regulation of body weight in mice under normoxic conditions. In the absence of HIF1a in neurons of mediobasal hypothalamus, the mice gained weight with increased food intake, accompanied with abnormal glucose and lipid metabolism. The hypothalamus senses and regulates many signals of biometabolic dynamics (16,17). In food-induced obesity, the whole-body energy homeostasis is disrupted (18), which may be related to abnormalities in hypothalamic neurons, e.g., NPY/ agouti gene-related peptidergic neurons stimulating POMC/ cocaine-and amphetamine-induced transcriptergic neurons (19)(20)(21). Conditional knockout of HIF in mouse hypothalamic POMC neurons using genetic methods and the knockout of HIF1b in the hypothalamic ARC have not revealed significant changes in feeding and body weight in mice fed with chow diet, but a significant increase in feeding and body weight is observed in those mice fed with HFD (7). In the present study, adeno- associated viruses containing syn (neuron-specific promoter) and cre fragments were injected into mediobasal hypothalamus of mice by stereotaxic injection, and HIF1a was conditionally knocked out in this site, confirmed by RT-PCR. Although we knocked out only HIF1a, rather than the HIF precursor, the body weight and food intake of mice with the chow diet increased significantly (p < 0.05), suggesting that neuronal HIF1a itself in mediobasal hypothalamus could be important for body weight and appetite control, although part of the reason may be due to broad regional HIF1a knockout neurons in the hypothalamus in our study. In addition to the hypothalamic POMC neurons and ARC region, HIF1a in other parts of the hypothalamic neurons may be involved in the regulation of body weight as well. In our results, the increase in body weight occurred earlier than the increase in food intake, suggesting that HIF1a in neurons of mediobasal hypothalamus might affect both the appetite and metabolic rate in mice. To exclude the effect of AAV-cre on body weight and food intake, we injected the same viruses in WT C57BL/6J mice. We did see the increase of body weight on 24 and 27 days after virus injection, but the increase of body weight was much less than HIF1a flox/flox mice. Moreover, the increase in food intake was not seen in WT mice. These results confirmed that the prominent weight gain and increased food intake in HIF1aKOMBH mice was due to loss of neuronal HIF1a in mediobasal hypothalamus. Studies found that the loss of HIF1a in the vascular endothelial cells of mice leads to a significant increase in the fasting glucose level, slow insulin response after intravenous glucose injection, delayed glucose clearance in the blood, and significantly impeded glucose absorption by the brain and heart (22). These results indicated that HIF1a plays an important role in glucose metabolism. Other studies showed that HIF induction can enhance hypothalamic glucose sensing (7), and the inhibition of hypothalamic HIF1 lead to glucose intolerance and increased serum insulin level (8). In our study, impaired glucose metabolism was present in HIF1aKOMBH mice. Although there were no significant changes in blood glucose levels in HIF1aKOMBH mice, the serum insulin levels of HIF1aKOMBH mice were significantly increased, suggesting that mice with neuronal HIF1a knockout in the mediobasal hypothalamus developed insulin resistance. In addition, the normal blood glucose levels might be the result of increased insulin secretion. The results of correlation analysis showed that the increase in insulin content was positively correlated with increased body weight in mice, indicating that the increase in insulin secretion might be the consequence of the increased body weight in mice.
IPGTT and insulin signaling results further confirmed abnormal glucose tolerance and insulin resistance in HIF1aKOMBH mice. The higher blood glucose levels in the knockout mice after glucose injection suggested that the mediobasal hypothalamic neuronal HIF1a knockout mice caused abnormal glucose tolerance in mice. The HIF1a KOMBH mice had a significantly delayed decrease in blood glucose after glucose injection, indicating that the absence of HIF1a in the neurons of medial basal hypothalamus might lead to impaired glucose clearance in the blood of mice. The glucose intolerance in HIF1aKOMBH mice may be due to higher insulin secretion baseline in these mice, which leads to insulin deficiency when facing glucose challenge. Akt, also known as PKB or Rac, plays a critical role in the control of survival and apoptosis (23)(24)(25). Insulin promotes a variety of important biological responses (26) and can stimulate the disposal of blood glucose primarily in target tissues, such as skeletal muscle and adipocytes, where the sugar is either oxidized or stored as glycogen or fatty acids. In both tissues, insulin promotes a rapid activation of specific PKB isoforms. Our results showed that the insulin-induced activation of AKT signaling pathway in the liver, epididymal fat, and skeletal muscle of HIF1aKOMBH mice was significantly lower than that of the control mice, confirming the reduced insulin sensitivity in these mice.
To investigate the lipid metabolism of HIF1a knockout mice in the mediobasal hypothalamus, we measured the serum TG, TC, FFA, HDL, and LDL levels in HIF1a flox/flox mice after virus injection. Although there are no significant changes in the serum TG content, the serum TC and FFA content in knockout mice significantly increased. The abnormal blood lipid content still appeared in the mice under chow diet, suggesting that HIF1a in the neurons in the mediobasal hypothalamus was also involved in blood lipid metabolism to some extent. The increased HDL and LDL level may be the consequence of increased lipid level.
To confirm that the HIF1a was indeed knocked out in HIF1aKOMBH mice, the RT-PCR of HIF1a was performed. When we checked the HIF1a in the whole hypothalamus, we found that the HIF1a did not decrease. Since the virus injections are mainly in the mediobasal hypothalamus, we took the ventral part of the hypothalamus to perform the RT-PCR again. As a result, the expression of HIF1a mRNA in HIF1aKOMBH mice decreased by 74%, indicating that AAV-hSyn-cre-GFP had a knockdown effect on HIF1a. Considering the knockout is neuronal specific, 74% decrease in HIF1a expression is reasonable.
The ARC in the hypothalamus has two distinct functional types of neurons that are important for appetite regulation. These neurons are related peptide agouti-related protein (AGRP)/NPY neurons, which express food-derived (appetite-stimulating) NPY and AGRP, and POMC neurons, which express POMC and inhibit appetite (27). NPY, as a polypeptide widely distributed in the central and peripheral nervous systems, plays an important role in body weight regulation. The main role of NPY is to increase food intake and reduce the thermogenic effect of satiated animals. The RT-PCR in the whole hypothalamus suggested nonchanged expression of POMC and NPY. Interestingly, however, in ventral hypothalamus, the mRNA expression of POMC and NPY were significantly reduced in HIF1aKOMBH group. The decreased POMC expression may explain the increased food intake in knockout mice, while the decrease in NPY mRNA expression might be the result of negative feedback regulation due to overeating in mice.
Previous study found that Glut4 is also expressed in the hypothalamus (28) and plays an important role in sensing glucose and regulating systemic glucose homeostasis. The knockdown of Glut4 in the mouse brain leads to impaired glucose tolerance, reduced insulin sensitivity, and impaired glucose-lowering regulation in mice (29). Moreover, the deletion of HIF1a in skeletal muscle cells leads to impaired Glut4 function (30). By contrast, the insulin receptor substrate 1associated phosphatidylinositol 3-phosphate kinase can activate Glut4 translocation to the plasma membrane, thereby importing glucose into the cell. Therefore, Glut4 is a key factor in skeletal muscle insulin sensitivity. The results of the present study showed a decreased Glut4 mRNA expression in HIF1aKOMBH mice in the whole hypothalamus, suggesting that Glut4 may be involved in the weight gain and abnormal glucose metabolism of these mice. However, the expression of Glut4 did not change in ventral hypothalamus, suggesting that the effect of HIF1a may act on the dorsal part of hypothalamus.
In summary, our study showed that neuronal HIF1a knockout at the mediobasal hypothalamus could lead to weight gain in mice accompanied with impaired glucose metabolism and lipid metabolism. POMC and Glut4 may be related to this effect of HIF1a. Whether the abnormal glucose metabolism and lipid metabolism was directly and causally related to the deletion of HIF1a at the mediobasal hypothalamus in mice remained unclear. In the future study, we will further investigate the effects of hypothalamic neuronal HIF1a on hypothalamic glucose sensing, insulin signaling pathway, and lipid regulatory pathway and mechanisms.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The animal study was reviewed and approved by Animal Ethics Committee of Xinjiang Medical University.

AUTHOR CONTRIBUTIONS
AR, WS, and QY performed experiments and analyzed data. AR and QY prepared the figures and drafted the manuscript. QY conceived and designed the research and approved the final version of manuscript. All authors contributed to the article and approved the submitted version.