Brown adipose tissue dysfunction promotes heart failure via a trimethylamine N-oxide-dependent mechanism

Low body temperature predicts a poor outcome in patients with heart failure, but the underlying pathological mechanisms and implications are largely unknown. Brown adipose tissue (BAT) was initially characterised as a thermogenic organ, and recent studies have suggested it plays a crucial role in maintaining systemic metabolic health. While these reports suggest a potential link between BAT and heart failure, the potential role of BAT dysfunction in heart failure has not been investigated. Here, we demonstrate that alteration of BAT function contributes to development of heart failure through disorientation in choline metabolism. Thoracic aortic constriction (TAC) or myocardial infarction (MI) reduced the thermogenic capacity of BAT in mice, leading to significant reduction of body temperature with cold exposure. BAT became hypoxic with TAC or MI, and hypoxic stress induced apoptosis of brown adipocytes. Enhancement of BAT function improved thermogenesis and cardiac function in TAC mice. Conversely, systolic function was impaired in a mouse model of genetic BAT dysfunction, in association with a low survival rate after TAC. Metabolomic analysis showed that reduced BAT thermogenesis was associated with elevation of plasma trimethylamine N-oxide (TMAO) levels. Administration of TMAO to mice led to significant reduction of phosphocreatine and ATP levels in cardiac tissue via suppression of mitochondrial complex IV activity. Genetic or pharmacological inhibition of flavin-containing monooxygenase reduced the plasma TMAO level in mice, and improved cardiac dysfunction in animals with left ventricular pressure overload. In patients with dilated cardiomyopathy, body temperature was low along with elevation of plasma choline and TMAO levels. These results suggest that maintenance of BAT homeostasis and reducing TMAO production could be potential next-generation therapies for heart failure.

The prognosis of severe heart failure (HF) remains unacceptably poor, and there is an urgent need to find a better treatment for this critical condition. Humans are considered to have around 6500 or more metabolites, and evidence indicates that alterations in the level of some metabolites have a close connection with heart failure 1 . Choline and trimethylamine-N-oxide (TMAO) are reported to increase in patients with heart failure and positively correlate with the severity of the New York Heart Association (NYHA) classification 2 . TMAO is also previously reported to promote atherosclerosis 3 . Choline in the diet is metabolized to trimethylamine (TMA) by gut flora and further oxidized into TMAO in liver 3 . Finally, TMAO enhances cholesterol accumulation in atherosclerotic plaque 3 ; however, the mechanistic link between TMAO and heart failure remains to be explored.
Brown adipose tissue (BAT) was initially characterized as a thermogenic organ, particularly in small rodents and human infants, but is now well known to be a metabolically active organ with a crucial role in maintaining systemic metabolic health in adult humans [4][5][6] . It was reported that a high-calorie diet induced impairment of BAT function in a murine model of obesity, leading to systemic glucose intolerance 7 . Systemic metabolic remodeling occurs in patients with heart failure, and low body temperature predicts a poor clinical outcome 8 . In a heart failure with preserved ejection fraction (HFpEF) murine model, BAT function was reported to be reduced 9 . While these reports suggest a potential link between BAT and heart failure, the potential role of BAT dysfunction in HF has not been fully investigated.
Here we show that BAT dysfunction develops with heart failure. This dysfunction led to increased TMAO levels in the circulation and heart. TMAO suppressed mitochondrial complex IV activity and reduced both ATP and phosphocreatine in cardiac tissues. In the advanced stage of heart failure, cardiac tissue becomes unable to utilize metabolites and enters a critical condition described as "out of fuel" 10 . Maintenance of BAT and inhibition of TMAO may be a potential therapy for heart failure.

Results
Left ventricular pressure overload reduces thermogenesis of brown adipose tissue. In agreement with the previous report 8 , we found that hospitalized patients with heart failure had a lower body temperature than a control group (Fig. 1A, Supplementary Fig. 1A,B). Because BAT has a critical role in maintaining body temperature 4-6 , we investigated how cardiac dysfunction could impact on BAT in two murine models of heart failure. In the first model, thoracic aortic constriction (TAC) was performed in wild-type (WT) mice at 11 weeks of age to generate left ventricular (LV) pressure overload as reported previously 11 . LV dysfunction developed 4 weeks after TAC ( Supplementary Fig. 1C-E) in association with significant reduction of both the body surface and intraperitoneal thermogenic responses ( Fig. 1B-D, Supplementary Fig. 1F). Body weight and food intake were comparable between the sham and TAC groups, but BAT became hypoxic with an increase of apoptotic cells, and BAT weight was significantly reduced after development of LV pressure overload (Fig. 1E-G and Supplementary Fig. 1G-K). Uncoupling protein-1 (UCP-1) is a proton channel in the inner mitochondrial membrane and uncouples the electron transport chain to generate heat instead of adenosine triphosphate (ATP) in BAT 12 . Cold exposure was reported to introduce inducible brown adipocyte like phenotype in white adipose tissue (WAT) 13 , and nowadays brown adipocyte like cells in white adipose tissues are described as beige cells 14 . We found that UCP-1 in BAT reduced with heart failure (Supplementary Fig. 2A), and level of beige markers remained low in subcutaneous white adipose tissue ( Supplementary Fig. 2B). To further characterize BAT in heart failure, we generated another murine model with heart failure. In the second heart failure model, cardiac dysfunction was induced by myocardial infarction (MI) at 11 weeks of age, also resulting in reduced BAT thermogenesis together with accumulation of apoptotic cells in this tissue 6 weeks after MI ( Supplementary  Fig. 2C-K). The results obtained in these two models suggested a close relation of heart failure to BAT dysfunction and impaired thermogenesis.
Next, we sought to generate a murine model with enhanced BAT function by the transplantation of donorderived BAT into intraperitoneal cavity of recipient mice. For this aim, WT mice were used as donor or recipient mice. BAT or WAT was transplanted into a visceral cavity of WT mice at 9 weeks of age, and TAC was performed at 11 weeks of age. At 2 weeks after TAC, mice were subjected to analyses. Transplantation of BAT improved  15,9). (B) Surface body temperature measured by a thermal camera in mice at 5 weeks after sham surgery (Sham) or TAC. (C) Acute cold tolerance test performed in mice at 4 weeks after Sham or TAC with measurement of body temperature in the scapular region (n = 6 and 7, respectively). (D) Hypothermia-free ratio during the acute cold tolerance test in mice at 4 weeks after Sham or TAC with measurement of the intraperitoneal temperature (n = 4, 4). (E) Pimonidazole staining of BAT in mice from (C) performed by the Hypoxyprobe-1 method. The right panel shows quantification of the hypoxic area (n = 4, 8). Scale bar = 50 μm. (F) Quantification of TUNEL-positive cells in BAT from mice at 6 weeks after Sham or TAC (n = 3, 3). (G) Body weight (BW)-adjusted BAT weight in mice prepared as described in Fig. 1C (n = 5, 4). (H) Hypothermiafree ratio during the acute cold tolerance test in mice with WAT or BAT transplantation at 2 weeks after TAC (n = 11, 17). (I) Assessment of cardiac function in mice with WAT or BAT transplantation at 2 weeks after Sham or TAC. FS: fractional shortening (n = 10, 11,9,19), LVDs: left-ventricular systolic dimension (n = 10, 11,9,19). Data were analysed by the 2-tailed Student's t-test (A, E, F, G and I), repeated measures followed by Tukey's multiple comparison test (C), or the log-rank test for Kaplan-Meier method (D, H). *P < 0.05, **P < 0.01. Values are shown as the mean ± s.e.m. Brown adipose tissue dysfunction deteriorates cardiac function after TAC . To further assess the role of altered BAT function in heart failure, we generated a BAT specific loss-of-function model. For this aim, we crossed Ucp1-Cre mice with Mfn1 flox/flox and Mfn2 flox/flox mice to obtain BAT Mfn double knockout (DKO) mice. Mitofusin1 and 2 (encoded by Mfn1 and Mfn2 respectively) are critically involved in the mitochondrial fusion process, and depletion of both Mfn1 and Mfn2 induced whitening of BAT ( Fig. 2A and Supplementary  Fig. 3A-C). TAC was performed in BAT Mfn DKO mice and their littermate control mice at 11 weeks of age, and mice were subjected to analyses at 1 week after TAC. After LV pressure overload was induced, BAT Mfn DKO mice showed a lower body temperature, higher mortality, and worse cardiac function than littermate control mice ( Fig. 2B-D). Apoptotic cardiomyocytes and cardiac fibrosis were also increased in BAT Mfn DKO mice (Fig. 2E,F). However, BAT Mfn DKO and littermate control mice showed comparable body weight, food intake, and cardiac function under basal conditions ( Supplementary Fig. 3D-F). In addition, we generated another BAT loss-of-function model by surgical removal of interscapular BAT (BATectomy). BAT was removed at 10 weeks of age, and TAC was performed at 11 weeks of age. At 2 weeks after TAC, mice were subjected to analyses. We confirmed the reproducible results in the BATectomy model ( Supplementary Fig. 3G-L). Chronic exposure to cold (15 °C) for 1 week at 2 weeks after TAC had no effect on the cardiac changes after TAC in BAT MfnDKO mice or WT mice, suggesting that modulation of environmental temperatures had minor effects on HF (Supplementary Fig. 4A-F).

Scientific
It is well accepted that sustained activation of the sympathetic nervous system promotes cardiac dysfunction 15,16 . Because we found an increase of norepinephrine in BAT after TAC ( Supplementary Fig. 4G), we examined the effects of chronic adrenergic stimulation on BAT function. We found that two-weeks infusion of isoproterenol significantly impaired thermogenic response to acute cold exposure ( Supplementary Fig. 4H) without inducing cardiac dysfunction ( Supplementary Fig. 4I), suggesting that chronic activation of adrenergic signaling during LV pressure overload induces BAT dysfunction which promoted cardiac dysfunction.
Choline and trimethylamine N-oxide are increased with heart failure. To clarify the mechanism by which alteration in BAT function promotes heart failure, we analyzed the levels of various metabolites in our models using CE-TOF/MS. Among cationic metabolites, we found an abundance of choline and choline-related metabolites in BAT (Fig. 3A) at 2 weeks after TAC (13 weeks of age). Notably, choline was increased in BAT after TAC, while phosphorylcholine was decreased (Fig. 3B). Metabolomic flux studies analyzing d9-choline showed that choline was incorporated by differentiated brown adipocytes and was metabolized further in the in vitro setting ( Supplementary Fig. 5A,B). At 4 weeks after transplantation of BAT or WAT, BAT transplantation without TAC led to a significant decrease of the plasma choline level (Fig. 3C). Recent studies have demonstrated that choline and its metabolites (including TMAO) are increased in patients with heart failure, showing a positive correlation with the severity of heart failure 2 . Consistent with previous reports, we found an increase of both plasma and cardiac TMAO levels 2 weeks after TAC (Fig. 3D,E). We also tested this in BATectomy model. BAT was removed at 10 weeks of age, and TAC was performed at 11 weeks of age. At 2 weeks after TAC, we analyzed plasma and heart samples and found that an increase of TMAO was enhanced in BATectomy with TAC than control with TAC ( Fig. 3F,G). These data suggested that circulating choline is taken up and metabolized by healthy BAT under physiological conditions, possibly to maintain the cell membrane integrity of brown adipocytes, while unprocessed choline is oxidized to TMAO after BAT dysfunction occurs.
Next, we determined the direct effect of TMAO on the failing heart. We continuously infused TMAO into WT mice for 2 weeks since 11 weeks of age and found that continuous infusion of TMAO led to a significant increase of the heart and plasma TMAO level (Fig. 3H). We also generated TAC model mice with TMAO infusion. We performed TAC operation in WT mice at 11 weeks of age. Two weeks after TAC, we started continuous infusion of TMAO and analyzed them additional 2 weeks later (totally 4 weeks after TAC). We found that TMAO treatment exacerbated cardiac dysfunction and fibrosis during pressure overload (Fig. 3I,J). In addition, plasma choline and TMAO levels were higher in patients with heart failure compared to controls (Fig. 3K).
It is reported that dietary choline is metabolized to trimethylamine (TMA) by gut microbiota, after which TMA is further metabolized to TMAO by flavin-containing monooxygenase (FMO) 17 . Therefore, we generated high choline-fed TAC mice model. At 2 weeks after TAC, WT mice were subjected to administration of a high-(1%) or low-choline (0%) diet for 2 weeks. Maintaining WT mice on a high-choline diet increased TMAO levels in plasma and myocardium, reduced cardiac function, and increased cardiac fibrosis (Supplementary Fig. 6A-G). These results suggest the contribution of increased TMAO during LV pressure overload in the progression of heart failure.
Inhibition of FMO ameliorates cardiac dysfunction during left ventricular pressure overload. We next examined the effects of FMO inhibition on heart failure using methimazole, a pan FMO inhibitor (Fmo-i) [18][19][20] . Fmo-i was administered WT mice since 11 weeks of age through drinking water for 1 week, and then choline infusion was performed. We found FMO inhibition effectively reduced circulating TMAO levels in the intravenous choline injection model (Fig. 4A). This treatment also decreased the plasma TMAO level after TAC together with improvement of cardiac dysfunction and fibrosis ( Fig. 4B-E).
There are five isoforms of flavin-containing monooxygenase in mammals, which are designated as Fmo1-5, and it has been reported that their expression varies among different cells and tissues as well as with gender and developmental stage 21,22 . The previous studies demonstrated that Fmo3 played a critical role in female mice, but male mice were known to have low Fmo3 expression 22 . Using bioinformatic analysis, we identified Fmo2 as a www.nature.com/scientificreports/ candidate enzyme for TMAO production in male mice. Thus, we generated a genetic model of systemic Fmo2 depletion in mice (Fmo2 KO mice), and performed TAC operation in these mice at 11 weeks of age. We found reduction of the cardiac TMAO level in these mice 2 weeks after TAC, along with improvement of cardiac function and fibrosis despite no change of heart weight ( Fig. 4F-I). These findings suggested that Fmo2 has a causal role for TMAO production in the presence of cardiac pressure overload and that inhibition of Fmo2 would be therapeutic target for heart failure. www.nature.com/scientificreports/ Trimethylamine N-oxide inhibits mitochondrial respiration in the heart. Next, we sought to elucidate the underlying mechanism of TMAO-induced cardiac dysfunction. We infused TMAO into 11 weeks old WT mice for 2 weeks. Administration of TMAO to WT mice significantly reduced cardiac tissue levels of ATP and phosphocreatine (Fig. 5A,B, Supplementary Fig. 7A). Electron microscopy of these mice revealed disruption of mitochondrial cristae in hearts obtained from mice with TMAO infusion and this was enhanced with LV pressure overload (Fig. 5C, Supplementary Fig. 7B). TAC led to similar pathological changes of cardiac mitochondria, and this mitochondrial disruption was augmented in BAT Mfn DKO and BATectomy models, and showed synergistic mitochondrial morphological alteration with TMAO administration in WT mice (Supplementary Fig. 7B-D). We also demonstrated that disarray of mitochondrial cristae in response to LV pressure overload was reduced in Fmo2 KO mice ( Supplementary Fig. 7E). Extracellular flux analysis of cardiac mitochondria isolated from WT mice with TMAO infusion showed reduction of oxidative phosphorylation and mitochondrial complex IV function (Fig. 5D).
To further investigate the underlying mechanism, we conducted RNA-sequencing (seq) analysis of cardiac tissues and proteomic analysis of cardiac mitochondria. We infused TMAO into WT mice at 11 weeks of age for 2 weeks, and then isolated cardiac mitochondria from these mice for RNA-seq or proteomics analyses. Our RNAseq analysis indicated the occurrence of metabolic remodeling in cardiac tissues ( Supplementary Fig. 8A), while the proteomic study identified cytochrome c oxidase subunit 1 (COXI) protein (also known as MT-CO1 and encoded by Mtco1) in cardiac mitochondria from control mice, but not in mice with TMAO infusion (Fig. 5E, Supplementary Fig. 8B). Western blot analysis and extracted ion chromatograms of two peptides unique to COXI protein revealed approximately 50% reduction of this protein by TMAO infusion (Fig. 5F). Quantitative PCR showed that Mtco1, Mtco2 and Mtco3 transcripts were not reduced by TMAO administration ( Supplementary  Fig. 8C), suggesting that TMAO inhibited mitochondrial respiration in the heart through the post-transcriptional regulation of mitochondrial complex IV protein.

Discussion
In the present study, we demonstrated that BAT dysfunction develops in murine models of heart failure and leads to abnormal choline metabolism. In turn, BAT dysfunction promotes heart failure through the production of the choline metabolite TMAO. Although Fmo3 was reported to have a critical role for TMAO production in female mice, it is well known that sex differences exist in the expression profiles of FMOs 22 . In this study, we adopted a bioinformatic approach and focused on Fmo2. Studies performed in male mice with genetic inhibition of Fmo2 showed that this enzyme has a critical role in converting choline to TMAO during heart failure, and revealed that TMAO suppresses cardiac metabolism by inhibiting mitochondrial complex IV. It is generally accepted that the intestinal flora converts dietary choline to TMA, after which TMA is metabolized to TMAO in the liver 17 . However, it remains to be determined how BAT dysfunction is linked to increased levels of TMAO and whether heart failure affects the activity of any FMOs.
Accumulating evidence indicates bacterial dysbiosis develops in patients with heart failure 23 . Pasini et al. reported that heart failure leads to an increase in pathogenic bacterial colonies including Candida, Campylobacter, Shigella, Salmonella, Yersinia enterocolitica 24 . Heart failure patients were also shown to have a decrease in Faecalibacterium prausnitzii and an increase in Ruminococcus gnavus and were associated with an increase in TMA-lyase, a key enzyme for TMAO generation 25 . Hayashi et al. showed positive correlations between the abundance of the genus Escherichia/Shigella and the level of TMAO 26 . Analyzing dysbiosis in murine heart failure models, testing the level of TMA-lyase, and showing the potential link among FMOs would be an interesting research topic to be explored.
We found FMO2 inhibition ameliorated cardiac dysfunction in mice. FMOs are known as regulators of stress resistance and play a role in xenobiotic-detoxifying and drug metabolism 27 . Endogenous function of FMOs are less clear, and it remains open question to be tested whether FMOs become a therapeutic target 27 . In the present study, we did not analyze the activity of BAT in patients with heart failure. Tahara et al. reported the results of [ 18 F]-fluorodeoxyglucose-positron emission tomography in a 23-year old female patient with heart failure who had low body temperature and suggested insufficient BAT-induced thermogenesis in this patient 28 . In our left ventricular pressure overload model with reduced ejection fraction, we found the diminished skin temperature at interscapular area where BAT predominantly exists. Thermogenic response under acute cold tolerance was also reduced. Under this condition, we found BAT becomes hypoxic together with an increase in apoptotic cells   www.nature.com/scientificreports/ on the role of BAT and abnormal choline metabolism in heart failure may lead to a better understanding of this complex condition (Fig. 5G). This study has several limitations. We have not provided clinical evidence showing that BAT dysfunction is associated with heart failure. BAT Mfn DKO mice were on a mixed background. Methimazole is not a specific inhibitor for FMOs. Only male mice were utilized for all experiments.

Methods
Human samples. Blood samples and clinical data were collected from patients of Niigata University Hospital. All subjects provided written informed consent prior to participation in these studies. The Scientific-Ethics Committees of Niigata University approved the protocols of all the studies (protocol number 2015-2292, 2017-0102), and each investigation was performed in accordance with the Declaration of Helsinki. Blood samples were immediately centrifuged to obtain plasma, which was subjected to CE-TOF/MS at Keio University. For body temperature studies, data were collected from admitted male Asian patients who were registered to our biobank between year 2012-2015. Analyses for the control group were done for patients who had EF > 50% and diagnosed as either (n = 3), arrhythmias (n = 10) or hypertension (n = 2). Analyses for the heart failure group were done for patients who were diagnosed as (n = 8) or dilated phase hypertrophic cardiomyopathy (n = 1). For metabolome studies analyzing TMAO or choline in plasma of control or heart failure patients, data for these were collected from metabolomic biobank generated with samples collected between year 2015-2018 in Niigata University. The control group included vasospastic angina (n = 10), paroxysmal supraventricular tachycardia (n = 6) and other cases including syncope with unknown reasons (n = 7). The heart failure group included idiopathic dilated cardiomyopathy (n = 6), dilated phase hypertrophic cardiomyopathy (n = 5), ischemic cardiomyopathy (n = 17) and other cases including amyloidosis (n = 2). Animal models. All animal experiments were conducted in compliance with the guidelines which was reviewed by the Institutional Animal Care and Use Committee of Niigata University and Juntendo University, and this study is approved by the Institutional Animal Care and Use Committee of Niigata University and Juntendo University. The study was carried out in compliance with the ARRIVE guidelines. Mice were housed in the animal facilities at Niigata University under specific pathogen-free conditions at a constant temperature of 23 °C, and a 12 h light/12 h dark cycle. Investigators were blinded to mouse genotypes during experiments. All mice were randomly allocated for surgical procedures under blinded condition. Before surgical procedures, mice are anesthetized with a cocktail of medetomidine hydrochloride (0.75 mg/kg, i.p.), midazolam (4 mg/kg, i.p.) and butorphanol tartrate (5 mg/kg, i.p.). TAC and MI were performed in 11-week-old male mice, as described previously 11 , and evaluation was done 4 weeks after surgery unless mentioned otherwise. Sham-operated mice underwent the same procedures, except for aortic constriction or coronary artery ligation. Transplantation of BAT or WAT was done as previously reported 29 . Surgical removal of BAT (BATectomy) was done as previously reported 30 . In some experiments, mice were housed at 15 °C (cold exposure) for 1 week with free access to water and chow under a 12 h light/12 h dark cycle. In some experiments, WT mice were fed a high-choline diet (10 g/kg, Dyets Inc., Bethlehem PA, USA) or low-choline diet (0 g/kg, Dyets Inc.). Wild-type mice were on a C57BL/6NCrSlc background. Mice expressing Cre recombinase in Ucp1-positive cells (Ucp1-Cre)(C57BL/6 J background) 31    Center and Harvard Medical School, Section on Integrative Physiology and Metabolism, Boston, USA) 31,33 . This cell line was established from wild-type FVB mice and was immortalized by infection with the pBabe retroviral vector encoding SV40T antigen. Cells were cultured in high glucose DMEM (Gibco, 12430) with 10% fetal bovine serum (FBS) and 100 U/ml penicillin/streptomycin solution (P/S), and differentiation was induced as described previously 34 . Fully differentiated brown adipocytes were used for analysis after 10 days of culture.
Acute cold exposure. Body temperature was assessed by subcutaneous implantation of a biocompatible sterile microchip transponder (IPTT-300 Extended Accuracy Calibration; Bio Medic Data Systems) in the scapular region or intraperitoneally according to the manufacturer's instructions. The cold tolerance test (CTT) was performed at 4 °C and body temperature of the animals was measured at hourly intervals for 8 h. Hypothermia was defined as a temperature < 35 °C in Supplementary Fig. 3I, < 32 °C in Fig. 1D or < 30 °C in Figs. 1H and 2C.
In some experiments, a tail tip cold exposure (TT-CE) test was performed. Before the TT-CE study, pelage hair on the backs of the mice was removed with a depilatory. The distal tails (approximately the distal 5 mm) were exposed to ice water for 10 min, and interscapular skin temperature was assessed with an infrared thermography camera (Testo, Inc., Sparta NJ, USA, 885) according to the manufacturer's instructions.

Scientific Reports
Enzyme prediction. Candidate enzymes with a role in choline metabolism could not be found by a homology search. Therefore, the E-zyme2 40 prediction tool, which was developed to predict the enzyme catalyzing an enzymatic reaction from the structures of two chemical compounds, was used to identify candidate enzymes for choline metabolism. E-zyme2 was employed after input of two KEGG compound identifiers, choline (C00114) and tma (C00565), and candidate enzymes were obtained.
Statistical analysis. Statistical analyses were done with SPSS version 24 software. All values were included in the analyses. If analyses did not reach statistical significance, in some cases outliers (shown as circles in figures) and abnormal values (shown as triangles in figures) were detected by SPSS boxplot analyses (boxplots show the upper whisker, upper quartile, median, lower quartile, and lower whisker) and excluded from further analyses (information described in the Excel raw data files). All outliers and abnormal values were also included in the Excel raw data files. If abnormal values were out of range, they were not shown in the figures and only included in the Excel format raw data files. Non-significant (NS) values in the figures indicate that these analyses included or excluded outlier and/or abnormal values and still did not reach statistical significance. All data are from different biological replicates, and are shown as the mean ± SEM. Differences between groups were examined by the two-tailed Student's t-test or two-way ANOVA, followed by Tukey's multiple comparison test, the non-parametric Kruskal Wallis test, or Dunnett's test for comparison among three or more groups. Survival curves were calculated by the Kaplan-Meier method and were compared with the log-rank test. Data from some experiments were analysed by 2-way repeated measures ANOVA, followed by Tukey's multiple comparison test. In all analyses, P < 0.05 was considered statistically significant.

Data availability
All data are available from the authors upon reasonable request. Additional material including source data is available online. Gene expression data obtained in these studies were deposited in the Gene Expression Omnibus database (GSE129756). The mass spectrometry proteomics data were deposited in the ProteomeXchange Consortium via jPOST 41 with the dataset identifier PXD013335.