Targeting miR-423-5p Reverses Exercise Training–Induced HCN4 Channel Remodeling and Sinus Bradycardia

Supplemental Digital Content is available in the text.

where C 2 is the plasma concentration, C 1 is the original plasma concentration, t 1/2 is the half time of the drugs (4 h) and t is elapsed time since the original injection (1 h). The effect of block of I f on the intrinsic heart rate was measured at this time. Heart rate was taken as an average over two 5-min periods (baseline) and over a single 5-min period (intrinsic heart rate; intrinsic heart rate+ivabradine). Heart rate variability was assessed in the following manner: continuous single lead ECGs were recorded digitally at 10 kHz using LabChart v7.0. 256 s ECG segments were selected after an appropriate acclimatisation period following each intervention (10 minutes preresting ECG, 5 min following intravenous atropine and propranolol, 60 min following oral ivabradine). R wave peaks were identified automatically and RR intervals were exported to Kubios v2.0 for analysis of heart rate variability. RR series were interpolated at 4 Hz. Heart rate variability was assessed in the time and frequency domains with the high frequency band defined as 0. 15-0.4 Hz.

Experimental animals
Care and use of laboratory animals conformed to the UK Animals (Scientific Procedures) Act 1986. Ethical approval for all experimental procedures was granted by the University of Manchester. Eight-week-old male C57BL/6J mice (Harlan Laboratories; initial body weight, 20-25 g) were randomly assigned to either sedentary or trained groups. Mice were housed five per cage in a temperature-controlled room (22°C) with a 12 h:12 h light:dark lighting regime and free access to food and water.

Swim training
Mice were subjected to a swimming programme described previously. 1,2 The mice were swimtrained for 60 min twice daily for 28 consecutive days. All mice were able to complete the course of training. Age-and weight-matched sedentary littermates served as controls for all experimental conditions and were handled daily. Additionally, a cohort of swim-trained mice was submitted to detraining for two weeks after the 28-day training period during which physical activity was restricted to the space of the cage.

Conscious ECG recordings
ECGs were recorded non-invasively in unrestrained, conscious mice using the ECGenie recording enclosure (Mouse Specifics, Inc., Boston, MA, USA) as described previously. 1 Heart rate was measured over 100 consecutive beats. The effect of (6 mg kg −1 ) ivabradine under complete autonomic block with atropine (0.5 mg kg −1 ) and propranolol (1 mg kg −1 ) was measured as previously described. 1 protocol. Amplification plots were analysed using RQ manager (Life Technologies). Ct values were exported to RealTime Statminer (Integromics) data analysis package that enabled advanced filtering of outlier genes, geNorm-based selection of optimal endogenous controls genes Gapdh and Tbp and differential expression testing using the non-parametric Limma test. 6 Transcript expression levels were calculated using the ΔCt method. Next generation sequencing Four cDNA libraries (two for sedentary and two for trained groups) were constructed from three pooled samples each. The cDNA libraries were prepared from 1 μg of total RNA using TruSeq Small RNA Sample Prep Kit (Illumina, Inc.) according to the manufacturer's instructions. Briefly, RNA 3′ adapter and RNA 5′ RNA adapter were ligated to each end of small RNA molecules.
The ligation products were used as a template for cDNA synthesis using SuperScript II Reverse Transcriptase (Invitrogen) to create single stranded cDNA. The cDNA was then PCR amplified using a common primer and a primer containing index sequences. After RT-PCR amplification, the cDNA libraries were purified by polyacrylamide gel electrophoresis to select libraries containing mature miR and other regulatory small RNAs. 22 and 30 nt bands were extracted from the gel and purified using the MinElute Gel Extraction Kit (Qiagen). The sizes of the selected library were validated by an Agilent Technologies 2100 Bioanalyzer using a High Sensitivity DNA chip. 50 base pair single-end reads were sequenced on the Illumina MiSeq sequencer (Illumina, Inc.) yielding up to 30 million raw reads per sample. Fastq files generated by MiSeq platform were analysed with FastQC (S. Andrews, 2010; FastQC: a quality control tool for high throughput sequence data; available online at: http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and any low quality reads and contaminated barcodes and primers trimmed with Trimmomactic. 7 Reads without adaptor sequences and with no ambiguous bases and the final trimmed length of at least 19 nucleotides were included for the final alignment analysis. Libraries were then aligned to mm10 assembly of mouse genome using Tophat 2 8 which incorporates Bowtie, short-read aligner software. Alignments with the best score were reported from each read. The mapped reads were then counted against gff files downloaded from miRbase, mmu.gff3, with HTSeq. 9 Reads were considered as mature miRs if they fulfilled the 'Strict' requirement on HTSeq, i.e. mapped to and within the whole miR range as defined by miRbase. Normalisation (to control for the variation in the number of read sequences across samples) was done using the DESeq Bioconductor package in 'R' based on the geometric mean. After normalised read counts were obtained, differentially expressed miRs were identified by comparing sedentary versus trained samples with DESeq. 10 DESeq is based on the negative binomial distribution and outputs fold change and P values for differential expression. P values were then adjusted for multiple testing using a false discovery rate of 5% via the Benjamini-Hochberg method. miRs with P<0.05 were considered to be differentially expressed.

Computational prediction of miR targets and cis-acting transcription factors
We used three established miR target prediction algorithms to investigate whether any differentially expressed miRs identified by next generation sequencing (FDR<0.05) were predicted targets in the regulation of mouse HCN4 based on 3′-UTR binding sites. miRs predicted by two out of three algorithms were considered targets for further analysis by reporter gene assay (miR-423-5p, miR-486-3p). In addition, we also carried out an unbiased search for all candidate miRs that could target HCN4 to find that it was a consistently predicted target for miR-27a-3p and hence this was included in the screen along with miR-1 that has been previously linked to HCN4. 1

Prediction algorithm
Predicted HCN4-targeting miRs MatInspector (Genomatix, Release 8.4) was used to analyse potential transcription factor binding sites within 2 kb of the 5 flanking region upstream of the transcription start site of the host gene of miR-423-5p, NSRP1. On the basis of evolutionary conservation between mice and humans and scores for similarity to canonical binding sites, 79 top predicted transcription factors were selected for further expression profiling along with nine other cardiac transcription factors either known to be involved in function and/or development of the heart (transcripts studied are listed in Table S4). (ii) pcDNA-NKX2.5. pEntr-Nkx2-5flbio containing mouse Nkx2.5 cDNA was a gift from William Pu (Addgene plasmid # 32969). The Nkx2.5 fragment was then transferred to the vector pcDNA6.2 cLumio-DEST (Invitrogen) by using the Gateway vector system (Invitrogen) to produce pcDNA6.2-Nkx2.5 using a protocol recommended by the manufacturer.
(iii) pGL3-NSRP1. A 2.1 kb fragment upstream of the NSRP1 transcription start site, corresponding to the promoter region and encompassing predicted Nkx2.5 binding sites (given in Figure S3) was synthesised by Dundee Cell Products. The fragment was then directionally subcloned into a luciferase containing plasmid, pGL3-basic (Promega) reporter using Kpn I and Hind III cloning sites, which were incorporated to the NSRP1 promoter construct, to generate a NRSP1 promoter luciferase construct. Cell culture, transfection and reporter assays H9c2 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen) supplemented with 10% foetal bovine serum and 1% penicillin/streptomycin. For Nkx2.5 overexpression, 5 x10 5 cells /well were plated in 6-well plates 24 h prior to transfection with 3 μg pcDNA3.1-Nkx2.5 or pcDNA3.1 empty vector. For reporter assays investigating miRs, cells were seeded at a density of 10 5 cells/well in 24-well plates 24 h prior to transfection and co-transfected with 500 ng HCN4 3-UTR plasmid or mutant and 0.5-1.5 μg precursor miR or negative control plasmid. For reporter assays testing NRSP1 promoter activity, the same procedures were followed to co-transfect H9C2 cells with 1 μg of promoter-luciferase fusion plasmid pGL3 and pcDNA-Nkx2.5 or negative control. Lipofectamine 2000 (Invitrogen) was used for all transfections according to the manufacturer's instructions. Transfected cells were incubated with DNA-Lipofectamine complexes for 24 h before lysing in passive lysis buffer (Promega) for luciferase assay or washed with phosphate buffered saline (PBS) and incubated for a further 24 h with 2 ml of fresh DMEM before lysis in Trizol (Invitrogen) for RNA extraction. Luciferase activity was determined using a Luciferase Assay System (Promega) using 10 μl of cell lysate on a luminometer (Berthold Technologies Lumat LB 9507). For each luminescence reading the injector was programmed to dispense 50 µl assay reagent after which there was a 2 s pre-measurement delay followed by a 7 s measurement period. Luciferase assays were performed in quadruplicate and repeated three times with an independent batch of cells. For miRs Firely luciferase and renilla luciferase activity were measured and data were analysed based on ratio of Firely/Renilla activity. Western blot HCN4 protein levels were determined by western blot using previously described methods. 14 Briefly, protein lysate was obtained by homogenising snap frozen sinus node biopsies using an MP FastPrep-24 5 G and 2 ml tubes containing FastPrep metal bead lysing matrix (1.4 mm) in RIPA buffer (Sigma Aldich). Total protein concentration was estimated using Bradford protein assay against standard curve of bovine serum albumin (BSA; 0-0.5 mg/ml) following which samples were denatured by adding final volume of 25% SDS-sample buffer -100 mM Tris-HCl, pH 6.8, 25% (v/v) glycerol, 10% (v/v) SDS, 10% (v/v) β-mercaptoethanol, 0.1% (w/v) bromophenol blue -and heating to 80˚C for 10 min. Samples were loaded onto a 12% stain-free SDS-polyacrylamide gel (Bio-Rad) with PreSciccion Plus (Bio-Rad) protein standards and run at 50 mV for ~50 min in SDS running buffer (25 mM Tris, 192 mM glycine, 0.1% SDS). Stain-free gels were imaged using ChemiDoc MP and then transferred to PVDF (polyvinyl difluoride) membranes using a Trans-Blot Turbo transfer system (Bio-Rad) at 15 V/0.3 mA for 15 min. PVDF membranes (activated in 100% ethanol before use) and thick filter paper were pre-wet in transfer buffer -1x Trans-Blot Turbo transfer buffer (Bio-Rad) and 20% (v/v) ethanol. Successful transfer was confirmed by using the ChemiDoc MP. PVDF membranes were washed in TBS for 5 min and then blocked in milk-TBS-Tween (5% w/v non-fat dried Marvel milk, 0.1% v/v TBS and Tween 20) for 1 h at room temperature with gentle rocking. The membranes were then probed with the following primary antibodies for 1 h at room temperature with gentle rocking: rabbit polyclonal anti-HCN4 (Alomone labs), 1:100; rabbit polyclonal anti-actin (Sigma Aldrich), 1:1000. Following three 10 min washes in TBS-Tween, membranes were then probed with horseradish peroxidase (HRP)-linked secondary antibody (HRP-linked anti-rabbit IgG, Cell Signalling) for a further 1 h at room temperature with gentle rocking. Membranes were then washed three times for 5 min in TBS-Tween to remove unbound secondary antibody. Chemiluminescence was achieved by the addition of Clarity Western ECL substrate (Bio-Rad) in a 1:1 ratio for 5 min in the dark. Membranes were then imaged with the ChemiDoc MP. Sedentary, trained and trained+antimiR samples were run on the same gel to ensure identical exposure conditions. The chemiluminescent signal intensity was normalised to the relative quantification of the corresponding intensity of actin. Data from each replicate were normalised and averaged across replicates. Proteomics Mass spectrometry based proteomics experiments were performed to evaluate protein expression of selected targets in isolated sinus node biopsies from sedentary male C57BL/6J mice (n=30 pooled into 3 samples with 10 biopsies in each sample). Briefly, sinus node biopsies were collected and immediately snap frozen in liquid N 2 and stored at -80°C until processing. Cardiac proteins were extracted from the biopsies and 1 mg protein from each sample was digested as described previously. 15,16 Peptides were desalted and fractionated by micro-flow reverse-phase ultra high pressure liquid chromatography into 12 fractions. Fractionated peptide samples were analyzed by online reversed-phase liquid chromatography coupled to a Q-Exactive Plus quadrupole Orbitrap tandem mass spectrometer. Peptide samples were separated on 15 cm fused-silica emitter columns using a 1 h multi-step linear gradient. Raw mass spectrometry data was processed using MaxQuant software (version 1.5.3.30) and proteins were identified with the built-in Andromeda search engine using a database containing all reviewed mouse SwissProt protein entries. Analysis of protein abundance was based on summed mass spectrometry-based protein intensities as determined by MaxQuant.

TUNEL assay
The right atrial wall including the sinus node of sedentary and trained mice were quickly dissected, flash frozen with liquid N 2 , and stored at -80°C until processing. The tissues were cryosectioned at 12 µm thick and mounted every five sections on an adhesion glass slide. Terminal deoxynucleotidyl transferase biotin-dUTP nick-end labelling (TUNEL; Roche, 11 684 795 910) was then performed combined with standard immunohistochemistry to label the sinus node using anti-HCN4 antibody. Briefly, frozen sections were fixed in 10% neutral buffered formalin, permeabilised in 0.1% triton X-100 in PBS, and blocked with 1% BSA, followed by incubation in rabbit polyclonal anti-HCN4 antibody (Alomone, APC-052, 1:200 dilution) at 4°C overnight. TUNEL was then carried out at 37°C for 60 min, according to the manufacturer's instruction. Cy3-conjugated anti-donkey and rabbit IgG secondary antibody (Merck Millopore, AP182C, 1:200 dilution) was added in the TUNEL reaction mix. As a positive control, sinus node sections were treated with micrococcal nuclease (ThermoFisher Scientific, EN0181, 30 U/ml). Images of HCN4-positive areas indicating the sinus node region were acquired using a laser scanning microscope (Zeiss LSM 5 PASCAL) equipped with a x40/1.0 PL Apo objective. The confocal settings were as follows: confocal aperture, 200 µm; scan speed, 1.60 µs pixel time, unidirectional; image size, 512 x 512 pixels. Images were acquired using the following conditions: 488 nm excitation and 505-530 nm emission for TUNEL, and 543 nm excitation and >560 nm emission for Cy3, respectively. To count the number of TUNEL-positive cells, images of HCN4-positive regions were combined and the contrast was enhanced using Adobe Photoshop. TUNEL positive cells per section were counted using Cell Counter of ImageJ at five different levels of the sinus node in five sedentary and trained mice.

Tissue electrophysiology
The beating rate of the isolated sinus node was determined by extracellular potential recording as described by Yamamoto et al. 17 In brief, animals were weighed and then killed by cervical dislocation following which a right atrial preparation encompassing the sinus node was rapidly dissected in Tyrode solution containing (in mM): 100 NaCl, 4 KCl, 1.2 MgSO 4 , 1.2 KH 2 PO 4 , 1.8 CaCl 2 , 25 NaHCO 3 and 10 glucose bubbled with 95% O 2 and 5% CO 2 to give a pH of 7.4. The preparation was superfused with 37°C Tyrode solution at a flow rate of 10 ml/min and extracellular potentials recorded using bipolar electrodes 100 μm in diameter. Recording electrodes interfaced with a Neurolog system (Digitimer) with low-pass and high-pass filters adjusted to optimise the signal-to-noise ratio. Extracellular potentials were continuously recorded for 20 min using a PC with a PowerLab and LabChart v7 software (ADInstruments) following which the effect of 2 mM CsCl (Sigma-Aldrich) on the beating rate was studied. The superfusing solution was changed to Tyrode solution containing 2 mM CsCl. After 15 min of treatment, the rate was recorded for 5 min. The preparation was then washed of CsCl for 20 min, during which the beating rate approached baseline values. The calculated rate was averaged over 500 beats. Intracellular action potential recording Intracellular action potentials were recorded in right atrial preparations containing the intact sinus node. Tissue was pinned to a specially designed chamber that allowed epicardial and endocardial contact with superfusing Tyrode solution (containing in mM: NaCl 120.3, KCl 4.0, CaCl 2 1.2, MgSO 4 1.3, NaH 2 PO 4 1.2, NaHCO 3 25.2 and glucose 11) bubbled with 95% O2 and 5% CO2 to give a pH of 7.4. Tyrode solution was circulated at 20 ml/min and tissues maintained at 37°C. The leading pacemaker site in the sinus node was mapped with bipolar extracellular electrodes as described in the preceding section. Using 3 M KCl filled sharp microelectrodes of 20-40 MΩ electrical resistance, intracellular action potentials were recorded at the leading pacemaker site of the sinus node and pectinate muscle (atrial tissue). Data acquired at 20 kHz was passed through a 10 kHz low-pass Bessel filter and amplified 10x by Axon Instruments GeneClamp 500 amplifier (Molecular Devices Inc), digitized with Axon Instruments Digidata 1440A (Molecular Devices Inc), and recorded onto a computer using the WinEDR v3.3.6 program (Dr. J. Dempster, University of Strathclyde, Glasgow, UK). Series of five consecutive action potentials were exported to LabChart v8 software (ADInstruments) and the following action potential parameters were measured: cycle length (interval between consecutive action potential peaks, ms), maximum diastolic potential (MDP, mV), maximum upstroke velocity (dV/dt max , mV/s), action potential height/amplitude (mV), action potential width (interval between consecutive maximum diastolic potentials, ms) and action potential duration (APD, ms) at 10, 50, 70 and 90% repolarization. Heart rate (beats per minute, bpm) was calculated from cycle length measurements for individual observations. GraphPad Prism 6 (GraphPad Software, Inc.) was used for statistical analysis.

Statistical analysis
Statistical analysis was carried out using GraphPad Prism 6 or 7 (GraphPad Software, Inc.) or SPSS (IBM). Two groups were analysed using an unpaired Student's t-test (two tailed). When the null hypothesis of equal variance was rejected, an unpaired t-test with Welch's correction was used. If the data were not normally distributed, a non-parametric test (Mann-Whitney test) was used instead of the unpaired t-test. To compare multiple groups, an ANOVA (one-or two-way) was used in the case of normally distributed data and the Kruskal-Wallis test in the case of data not normally distributed. P<0.05 was regarded as significant. 0.12>P<0.05 was regarded of potential interest and the precise P value is given. In figures, data are shown as meansSEM; asterisks indicate significance. For TLDA cards, a non-parametic Limma test was used to compare differences between sedentary and control animals. 6 Statistical analysis of the next generation sequencing data is described above.

SUPPLEMENTAL DISCUSSION
Intrinsic heart rate of untrained, young, male, human subjects Jose and Collinson 19 reported the intrinsic heart rate for 139 male subjects (non-athletes) between the ages of 20 and 30. This is the first and largest study of the intrinsic heart rate. We have data for 10 male subjects (non-athletes) between the ages of 20 and 30. Although the intrinsic heart rate in this study (97.92.6 beats/min) is statistically different (Student's t test; P=0.002) from the intrinsic heart rate (105.60.6 beats/min) reported by Jose and Collinson, 19 it is within the mean  2 standard deviations from the study of Jose and Collinson. 19 The mean  2 standard deviations encompasses 95.4% of the data from the Jose and Collinson 19 study. Previously, we have put this forwards as a criterion for acceptance of intrinsic heart rate data. 20 Whereas the intrinsic heart rate of untrained, young, male, human subjects falls within this acceptable range in some studies, lower values (the lowest being 83.1 beats/min) have been reported in other studies; 20 in all these studies the number of human subjects was 10 or less and, in this respect, have to be considered less definitive than the study of Jose and Collinson. 19 The intrinsic heart rate of young, male subjects (non-athletes) should be approximately the same in different studies. The low intrinsic heart rates reported in some studies are likely to be the result of a technical issue, because the only factors known to decrease the intrinsic heart rate (age, 19 heart failure 21 and athletic training e.g.22 ) are unlikely to apply.

Reported evidence of exercise training-induced increase of vagal tone
Based on block of autonomic tone using atropine and propranolol, we have found no evidence of altered autonomic tone and in particular high vagal tone in human athletes (this study; Figure 1B) and exercise trained C57BL/6J mice. 1 However, again based on block of autonomic tone using atropine and propranolol, Guasch et al. 23 (using the Wistar rat) and Aschar-Sobbi et al. 24 (using the CD1 mouse) have recently reported evidence of high vagal tone following exercise training (although, strangely, Guasch et al. 23 stated that there was no exercise training-induced bradycardia in their study). The cause of the discrepancy is unknown, although it could be the result of an unknown technical issue, the doses of atropine and propranolol used, species used, strain of mouse used, and differences in the duration, intensity and type of training.
We have reviewed the role of high vagal tone (assessed by pharmacological block of autonomic tone) in exercise training-induced bradycardia in the human and animal models. 20 In all studies of human athletes in which the measurement of the intrinsic heart rate is deemed to be correct (see above) there is no evidence of high vagal tone. 20 In nine animal studies, the data suggests that high vagal tone accounts for 76% (mouse), 40% (rat), 43.6% (rat), 10% (rat), 0% (rat), 0% (rat), 0% (rat), 0% (rat) and 0% (rat) of the exercise training-induced bradycardia. 20 Aschar-Sobbi et al. 24 state that high vagal tone accounts for 100% of the exercise training-induced bradycardia in the mouse. This is a higher contribution than any of the previous studies. In contrast, in the mouse we have argued that its contribution is 0%, 1 consistent with five of the previous studies. Also in our case, we have based our conclusion not only on autonomic blockade in vivo in the mouse; we have also based it on intrinsic heart rate measurements from the isolated sinus node from both the rat and mouse. 1 Furthermore, we have shown that there is an exercise training-induced downregulation of HCN4 mRNA, HCN4 protein and funny current in the rat and mouse and the exercise training-induced bradycardia is abolished on block of funny current in the mouse (this study and D'Souza et al. 1 ).
We conclude that it is likely that downregulation of funny current is playing a role in exercise training-induced bradycardia. However, the evidence from Guasch et al. 23 and Aschar-Sobbi et al. 24 of exercise training-induced high vagal tone is difficult to refute and perhaps high vagal tone could play a role in some circumstances.

Nkx2.5 expression in the adult sinus node
While much is known about the regulatory networks at play in the embryonic development of the cardiac conduction system, the transcriptional networks maintaining function of the adult sinus node are comparatively understudied. The data presented in Online Table IV, as far as we are aware, is the first large scale transcriptomic analysis of transcription factors within the adult mouse sinus node. In Online Figure XA, data from Online Table IV are plotted to show expression levels of 88 transcription factors in the sedentary adult mouse sinus node. Transcriptionally, Nkx2.5 (red bar) is the ninth most abundant transcription factor (of the transcription factors measured). Expression levels of well-known transcription factors in the sinus node (Tbx3, Tbx18 and Shox2) are highlighted in blue for comparison. Online Figure XB shows the protein expression level of selected transcription factors (measured by mass spectrometry) in the sedentary adult mouse sinus node. This confirms the presence of Nkx2.5 in the adult mouse sinus node. Finally, in recent work, Wu et al. 25 found that H3K4me3 modifications (a prominent active histone mark associated with active genes) were highly enriched in the Nkx2.5 promoter in the mouse sinus node. In summary, these observations suggest that there is baseline expression of Nkx2.5 in the sinus node (which is then increased in response to exercise training).  Table I. Characteristics of human subjects. All athletes participated in endurance sports (running, n=4; triathalon, n=3; cycling n=1). BMI, body mass index; SDNN, standard deviation of normal to normal beats; cSDNN, heart rate corrected SDNN.  Table II. Intracellular action potential parameters from control mice (sinus node from 4 mice and right atrium from 3 mice) and exercise trained mice (sinus node from 8 mice and right atrium from 5 mice). APD 10 etc., action potential duration at 10% repolarization etc. Significant differences highlighted by P values in bold font.  Online Figure VII. The antimiR has little or no effect on heart weight:body weight ratio, PR interval, QRS duration and QTc interval of sedentary or trained mice. n=5/5/5/6 for heart weight:body weight ratio. ECG parameters measured under isofluorane anaesthesia. Continuous 100 beat recordings were analysed and averaged. n=10/10/5/11 for ECG parameters. *significantly different (P<0.05).

Control
Online Figure X. Expression of Nkx2.5 in the sedentary adult mouse sinus node. A, mRNA expression level of 88 transcription factors in the sedentary adult mouse sinus node (n=8) measured using Taqman low density array cards and normalised to expression level of housekeeping genes GAPDH and TBP. Nkx2.5 is highlighted in red and, for comparison, Tbx3, Tbx18 and Shox2 are highlighted in blue. B, Protein expression level (on logarithmic scale) of selected transcription factors, including Nkx2.5, in the sedentary adult mouse sinus node (n=3 cohorts of 10 mice) measured using high-resolution mass spectrometry. Protein expression level of HCN4 also shown. Indivdual data points (for the three cohorts of mice) as well as means±SEM shown. MS, mass spectrometry. SN, sinus node.