Exercise-induced activation of ventromedial hypothalamic steroidogenic factor-1 neurons mediates improvements in endurance

SUMMARY

Repeated exercise produces robust physiological benefits and is the leading lifestyle intervention for human health. The benefits from exercise training result from the remodeling of skeletomuscular, cardiovascular, metabolic, and endocrine systems. In mice, we find that activation of the central nervous system following exercise is essential for subsequent endurance performance and metabolism benefits. Ventromedial hypothalamic steroidogenic factor-1 (SF1) neurons are activated following exercise and repeated training results in increased post-exercise SF1 neuron activation. Exercise training increases the intrinsic excitability and density of excitatory synapses on SF1 neurons, suggesting that exercise history is encoded through hypothalamic plasticity. Inhibition of SF1 neuron output blocks endurance gains and metabolic improvements that result from exercise training. Conversely, stimulation of SF1 neurons following exercise enhances gains in endurance. These results demonstrate that exercise-induced hypothalamic SF1 neuron activity is essential for the coordination of physiological improvements following exercise training.

Graphical Abstract

IN BRIEF

Kindel et al. show that repeated exercise potentiates the activity of ventromedial hypothalamic SF1-expressing neurons. SF1 neural activity is required for mice to achieve the endurance benefits of exercise and stimulation of SF1 neurons following training enhances endurance. This highlights the CNS as a critical component mediating adaptations to exercise.

INTRODUCTION

Physical exercise engages adaptive mechanisms that facilitate future exercise performance (colloquially referred to as endurance) and improves physiological function.^1,2 The mechanisms underlying improvements in endurance have traditionally been viewed as processes that occur in the periphery, including remodeling of the skeletomuscular,^3–5 cardiovascular,^6,7 metabolic,^8–10 immune,^11,12 and endocrine systems.^13,14 Although exercise is associated with remodeling of brain circuits, including increased rates of neurogenesis,^15–17 increased synaptic connectivity,^18–22 and reduced neuroinflammation,^23–25 these adaptations are presumed to reflect, rather than produce, the endurance phenotypes associated with repeated exercise.

Excitatory neurons in the ventromedial hypothalamus (VMH) express the orphan nuclear receptor and transcription factor steroidogenic factor-1 (SF1).²⁶ SF1 expression is required in development for the proper differentiation of VMH neurons^27–29 and maintenance of energy homeostasis in adulthood.^30–33 SF1-expressing VMH neurons integrate peripheral signals such as leptin, insulin, and glucose to regulate energy expenditure.^34–36 Additionally, genetic strategies that reduce central SF1 expression impair exercise-induced increases in endurance capacity,³⁷ suggesting a critical role for this factor in the adaptive response of VMH neurons to metabolic stressors. However, the role of SF1-expressing VMH neurons in mediating the adaptive physiological and metabolic responses to exercise remains unknown.

Here, we demonstrate a critical role for VMH SF1 neurons in mediating adaptations to exercise. We find that SF1 neuron activity is essential for appropriately engaging energy stores, activating skeletomuscular remodeling during exercise, and enhancing endurance with repeated training. We demonstrate that activity increases in a subset of VMH SF1 neurons after exercise. Repeated exercise increases both the number of VMH SF1 neurons activated and the magnitude of their activation. With repeated exercise, SF1 neurons show increased intrinsic excitability and a greater number of excitatory synaptic inputs. Inhibiting SF1 neural activity following exercise training blocks subsequent improvements in endurance capacity. Conversely, exogenous activation of SF1 neurons immediately following exercise robustly enhances endurance performance, demonstrating that SF1 neural activation initiates metabolic adaptations to exercise. Collectively, these results identify VMH SF1 neurons as a critical central nervous system node for controlling metabolic adaptations to exercise.

RESULTS

VMH SF1 output is required for exercise-induced adaptations and endurance improvements.

Because VMH development influences exercise improvement,³⁷ we first sought to determine whether VMH neurons are activated by exercise. Following a single exercise session, we found an upregulation of the activity-dependent brain derived neurotrophic factor (Bdnf) transcript.^38,39 Exercise increased the percentage of neurons expressing Bdnf and the amount of Bdnf transcripts in VMH SF1 neurons (Figures 1A–D and S1A–D).

Figure 1. VMH SF1 output is required for physiological adaptations to exercise.

(A) In situ hybridization strategy: after 3 days of habituation, mice completed a single exercise session (n = 6 mice) or remained sedentary (n = 4 mice). In situ hybridization was performed on tissue containing the VMH with probes for Sf1 and Bdnf.

(B) Sf1 (cyan) and Bdnf (magenta) expression in the VMH of sedentary (top) and exercised (bottom) mice. DM/C, dorsomedial/central; VL, ventrolateral. Scale = 200 μm (left and middle), 50 μm (right).

(C) Proportion of cells expressing Bdnf in the VMH DM/C division in sedentary and exercised mice.

(D) Number of Bdnf transcripts per SF1-expressing neuron in the DM/C division of the VMH in sedentary (n = 1873 cells) and exercised (n = 4065 cells) mice.

(E) SF1 neuron inhibition strategy (top): SF1-Cre mice were injected with a Cre-dependent (DIO) TetTox virus (n = 10 mice). Cre-negative littermates injected with the same virus served as controls (n = 10 mice). Exercise stress test (bottom): mice were run on a stress test paradigm until exhaustion while indirect calorimetry was performed.

(F) VO₂ max of untrained SF1:TetTox and control mice in the exercise stress test.

(G) Distance run prior to exhaustion on the exercise stress test in SF1:TetTox and control mice.

(H) Respiratory exchange ratio (RER) aligned to the start of the exercise stress test in SF1:TetTox and control mice.

(I) RER at exhaustion in SF1:TetTox and control mice.

(J) Muscle transcriptomics after exercise in trained SF1:TetTox (n = 4 mice) and trained control (n = 5 mice) mice, relative to sedentary control (n = 5 mice) mice. Quadriceps muscle was removed after an exercise session and bulk RNAseq was performed. The results showed a number of differentially expressed genes (DEGs) in control (top) and SF1:TetTox (bottom), relative to sedentary control group.

(K) Volcano plot of DEGs in the quadriceps muscle 3 hours after exercise.

(L) SF1:TetTox (n = 5 mice) and Cre-negative littermate control (n = 6 mice) mice performed a three-week training paradigm with endurance tests run each week.

(M) Change in endurance distance from baseline (week 0) to week 3 for SF1:TetTox and control mice.

Error bars and shading represent SEM. Dots represent individual mice. Two-way ANOVA main effect of group: ☼☼☼ p < 0.001; two-way ANOVA group x time interaction: ††† p < 0.001. Tukey post-hoc, Mann-Whitney, and unpaired t-tests: **p < 0.01, ***p < 0.001.

See also Figure S1.

Given this exercise-induced upregulation of an activity-dependent gene, we explored the possibility that neural activity in VMH SF1 neurons is necessary for the physiological adaptations that occur with exercise. We expressed a Cre-dependent tetanus toxin light chain (TetTox) in the VMH of SF1-Cre mice to prevent synaptic vesicle exocytosis (Figure 1E).⁴⁰ In the absence of VMH SF1 neuronal transmission, we found that mice had similar VO₂ max (Figure 1F) but diminished endurance and running speeds during an exercise ‘stress test’ (Figures 1G and S1E–G). SF1:TetTox-expressing mice preferentially oxidized carbohydrates at lower exercise intensities with a faster shift from lipid to carbohydrate oxidation than in control mice, despite similar oxygen consumption (Figures 1H–I and S1G). These results suggest that SF1 neural activity influences fuel utilization during exercise.

Regular, submaximal exercise training induces a cascade of training-dependent transcriptional changes in skeletal muscle. These changes improve fuel utilization efficiency and mitochondrial function,^37,41 which enable increased endurance capacity. We found that normal exercise-dependent gene expression changes in skeletal muscle were nearly abolished in trained SF1:TetTox mice following a run to exhaustion (Figures 1J and 1K). Given the established role for these molecular adaptations in endurance capacity,^37,41 we reasoned that impairing SF1 neuronal activity might also prevent the normal training-dependent increase in endurance capacity. We found that VMH SF1:TetTox-expression prevented improvements in endurance capacity in mice that underwent regular treadmill training sessions (Figures 1L, 1M, and S1J–N). Similarly, SF1:TetTox mice did not increase voluntary exercise when given free access to a running wheel (Figures S1H and S1I). The inability to improve endurance capacity was not simply a consequence of poor initial performance or body weight. Mice with SF1 neuron-specific knockdown of BDNF—which also impaired initial performance and led to increased body weight—had rapid improvements in endurance after just one week of training (Figures S1O–X).

Taken together, these experiments suggest that the exercise-related output of VMH SF1 neurons encodes a prominent signal required for adaptive metabolic responses associated with exercise and endurance.

VMH SF1 neural activity is potentiated by exercise training.

Because the output of VMH SF1 neurons is necessary to encode the physiological adaptations that lead to improvements in endurance capacity, we hypothesized that exercise would activate SF1 neurons. To test this hypothesis, we used a genetically encoded calcium indicator (GCaMP6s) to monitor Ca²⁺ dynamics of SF1 neurons before, during, and after treadmill exercise (Figures 2A and S2A). We observed increases in VMH SF1 population Ca²⁺ dynamics at the onset and termination of endurance exercise sessions (Figures S2B, S2C, S2E, and S2F), along with a sustained elevation of activity following exercise (Figures S2D and S2G).

Figure 2. Exercise training increases SF1 neuron activation.

(A) Endoscopic (miniscope) imaging strategy: Cre-dependent GCaMP was expressed in the VMH of SF1-Cre mice (n = 93 neurons, 4 mice), and a GRIN lens was implanted above the injection site. Calcium dynamics during endurance sessions were recorded before and after a 1-week training period. Recording sessions consisted of a 2.5-min baseline, an endurance session, and a 5-min recovery.

(B) VMH SF1 calcium dynamics from individual neurons aligned to the end of exercise on day 1 and on day 8. Neurons are arranged by time of peak activity and classified into either post-run activated (top, purple) or post-run inhibited (bottom, blue). Each row is one cell.

(C) Distribution of neurons that were activated, inhibited, or had no change in activity at the end of exercise on day 1 (left) and day 8 (right) endurance recording sessions.

(D) VMH SF1 calcium dynamics at the end of exercise before and after training in post-run activated neurons.

(E) Mean GCaMP activity in the 5 min after the end of exercise in post-run activated neurons in the recording session on day 1 and day 8.

(F) VMH SF1 calcium dynamics at the end of exercise before and after training in post-run inhibited neurons.

(G) Mean GCaMP activity in the 5 min after the end of exercise in post-run inhibited neurons in the recording session on day 1 and day 8.

(H) Example of neurons tracked between day 1 and day 8 (left). Activity at the end of exercise from individual neurons tracked across day 1 and day 8 (right). Examples of each possible outcome of neurons in relation to their post-run activated (A) and inhibited (I) categorization from day 1 to day 8.

(I) Sankey diagram of SF1 neuron activity classification on day 1 and day 8.

(J-M) Mean calcium activity in the 5 min after the end of exercise of tracked neurons (J) activated at the end of exercise in both sessions (A → A); (K) inhibited at the end of exercise activated in the first session and activated in the second (I → A); (L) activated at the end of exercise in the first session and inhibited in the second (A → I); (M) inhibited at the end of exercise in both sessions (I → I).

Error bars and shading represent SEM. Lines and dots represent individual neurons. Paired and unpaired t-tests: *p < 0.05, **p < 0.01, ***p < 0.001.

See also Figures S2 and S3.

To understand how exercise training changes individual SF1 neuron activity, we performed longitudinal endoscopic Ca²⁺ imaging of SF1 neurons before and after a week of treadmill training (Figures 2A, S2H, and S2I). During the initial exercise session, we observed two distinct groups of exercise-responsive VMH SF1 neurons. One group was active at various phases but inhibited at exercise termination (‘post-run inhibited’), while the other group was activated upon exercise termination (‘post-run activated’). These exercise activated SF1 neurons are distinct from neurons activated during social interactions (Figure S3). While a week of exercise training did not substantially change SF1 neural activity at the beginning of exercise (Figures S2J–M), we observed that a greater number of post-run activated SF1 neurons on day 8 compared to day 1 (Figures 2B and 2C). The magnitude of this activity was also significantly higher (Figures 2D–G). Tracking individual neurons across sessions⁴² confirmed these shifts in neural activation (Figures 2H–M).

Increased excitatory drive onto SF1 neurons following repeated exercise training.

Our results demonstrate that training increases the number of post-run activated SF1 neurons and potentiates the magnitude of their activity. To determine the physiological mechanisms underlying this plasticity, we made whole-cell patch clamp recordings of VMH SF1 neurons in acute brain slices from mice that underwent three weeks of exercise training (Figure 3A). Relative to neurons from sedentary mice, VMH SF1 neurons from exercised mice exhibited a depolarized resting membrane potential and increased frequency of spontaneous action potentials (Figures 3B–D). Changes in resting membrane potential were observed from both animals trained on a treadmill (Figure 3C) and from those partaking in voluntary exercise (Figures S4A–D). In acute brain slices made after an exercise session, we observed a robust increase in action potentials in SF1 neurons of trained mice (Figure 3B). Specifically, SF1 neurons from sedentary mice had a mean AP frequency of 2.4 ± 0.6 Hz with 36.9% (7/19) of SF1 neurons effectively silent (spike rates < 0.5 Hz^43,44). In contrast, SF1 neurons from exercised mice had mean AP frequency of 6.8 ± 1.3 Hz and we did not observe any silent SF1 neurons (0/12) in slices from exercised mice (Figure 3D). These changes in spiking activity were accompanied by an approximately 100% increase in the frequency of excitatory postsynaptic currents (EPSCs) with repeated exercise relative to VMH SF1 neurons recorded from sedentary mice (Figures 3E and 3F). EPSC amplitudes remained unchanged in VMH SF1 neurons across the two conditions (Figure 3G). The frequency change was specific for excitatory synaptic currents, as we found no change in the frequency or amplitude of inhibitory postsynaptic currents in the same cells (Figures S4E–H).

Figure 3. Exercise training increases excitatory input to VMH SF1 neurons.

(A) Slice electrophysiology: SF1 neurons were identified by tdTomato reporter expression from sedentary (n = 2 mice) and exercised (n = 3 mice) SF1-Cre::Ai9 mice trained for four weeks. On the day of the recording, exercised mice were run on an endurance session, acute sections of the VMH were prepared, and whole-cell patch clamp recordings were made from tdTomato-expressing neurons.

(B) Example voltage traces of VMH SF1 neurons from sedentary (top) and exercised (bottom) mice.

(C) Resting membrane potential of VMH SF1 neurons from sedentary (n = 19 neurons) and exercised (n = 12 neurons) mice.

(D) Action potential frequency of VMH SF1 neurons from sedentary and exercised mice.

(E) Example excitatory postsynaptic current (EPSC) traces of VMH SF1 neurons from sedentary (top) and exercised (bottom) mice.

(F) EPSC frequency of VMH SF1 neurons from sedentary (n = 18 neurons) and exercised (n = 11 neurons) mice.

(G) Mean EPSC amplitude of VMH SF1 neurons from sedentary and exercised mice.

(H) Synaptic input: Cre-dependent GCaMP was expressed to visualize VMH SF1 neurons from sedentary (n = 5 mice) and exercised (n = 5 mice) SF1-Cre mice.

(I) Example dendritic spines on VMH SF1 neuron dendrites from sedentary (top) and exercised (bottom) mice. Scale = 5 μm.

(J) Dendritic spine density on VMH SF1 neuron dendrites from sedentary (n = 31 spines) and exercised (n = 26 spines) mice.

Error bars represent SEM. Dots represent individual cells or dendrites. Unpaired t-tests: **p < 0.01, ***p < 0.001.

See also Figure S4.

The increased frequency of EPSCs found on neurons from exercised mice suggested exercise might increase the number excitatory synapses on SF1 dendrites. VMH neurons receive excitatory inputs on dendritic spines, with previous work demonstrating that the density of dendritic spines on VMH neurons is plastic.⁴⁵ We examined dendritic spine density and morphology in sparsely labeled VMH SF1 neurons (Figures 3H). Consistent with the change in EPSC frequency, we found that repeated exercise doubled the dendritic spine density on VMH dendrites (Figure 3I and 3J). This effect was independent of branch surface area (Figure S4I–K), indicating that new spines were formed on both thick and thin diameter dendritic branches. To demonstrate that these spines contained synaptic proteins, we assessed the presence of Homer1, a protein known to be localized to postsynaptic densities of excitatory synapses.^46,47 We detected Homer1 in a similar fraction of spines in both sedentary and exercise conditions, suggesting that the increase in spine number leads to an increase in functional synapses (Figures S4L–O). Taken together, these results suggest that exercise training builds excitatory drive to VMH SF1 neurons.

SF1 neuron activity following exercise enhances endurance capacity.

The increased activity at the termination of exercise suggested that VMH SF1 neuron signaling immediately after exercise training may be the critical signal driving training-induced improvements in endurance capacity. To directly determine the role of VMH SF1 neural activity after exercise, we used bidirectional optogenetic manipulations to either acutely inhibit or enhance VMH SF1 neuronal activity following daily exercise training. We found that optogenetic inhibition of SF1 activity at the end of each training session blunted the improvements in endurance normally observed with exercise training (Figures 4A–F) and prevented the normal post-exercise increase in blood glucose without affecting body weight (Figures S5A–D). Conversely, optogenetic enhancement of post-exercise SF1 activity increased endurance capacity as animals reached performance plateaus (Figures 4G–L and S5E–H), suggesting that further increasing post-run SF1 neuron activity can enhance the adaptations driven by exercise. Besides the well documented effect of increasing blood glucose (Figure S5G),⁴⁸ SF1 stimulation alone increased relative carbohydrate utilization (Figures S5I–K), overall energy expenditure (Figure S5L), oxygen consumption (Figures S5M and S5N), and locomotion (Figures S5O and S5P); however, the increased activity represents a small fraction of the exercise training load (Figure S5O). These findings suggest that SF1 neural activity after exercise has a potent role in increasing endurance capacity.

Figure 4. SF1 neural activity after exercise modulates subsequent endurance capacity.

(A) Optogenetic inhibition strategy: Cre-dependent StGTACR2 was bilaterally injected into the VMH of SF1-Cre mice (n = 6 mice) and Cre-negative littermate control mice (n = 6 mice). Fiber optics were implanted over the injection sites.

(B) Training protocol: Mice underwent three weeks of training with an endurance test run once per week. VMH SF1 neurons were inhibited for 15 minutes after each training session.

(C) Change in endurance distance from baseline across three weeks of training in Cre− control and Cre+ mice.

(D) Change in endurance distance from baseline to week 1.

(E) Maximum treadmill speed in Cre− control and Cre+ mice in week 1 endurance test.

(F) Work performed in Cre− and Cre+ mice in week 1 endurance test.

(G) Optogenetic excitation strategy: Cre-dependent ChR2 (n = 12 mice) or a control eYFP fluorophore (n = 12 mice) was injected into the VMH of SF1-Cre mice and a fiber optic implanted over the injection site.

(H) Training protocol: Mice underwent three weeks of training with an endurance test run once per week. VMH SF1 neurons were stimulated for 60 min after each training session.

(I) Change in endurance distance relative to baseline (week 0) across three weeks of training in eYFP and ChR2 mice.

(J) Change in distance achieved from the final endurance test relative to baseline in eYFP and ChR2 mice.

(K) Maximum speed achieved on the final endurance test in eYFP and ChR2 mice. Speed was not increased beyond 30 m/min.

(L) Work performed in the final endurance test in eYFP and ChR2 mice.

Error bars and shading represent SEM. Dots represent individual mice. Two-way ANOVA main effect of group: ☼ p < 0.05, ☼☼ p < 0.01; two-way ANOVA group x time interaction: † p < 0.05; Tukey post-hoc and unpaired t-tests: *p < 0.05, **p < 0.01.

See also Figure S5.

Taken together, these data suggest a prominent role for VMH SF1 neuron activity in controlling body-wide responses to repeated exercise⁴⁹ and suggest a central role for the brain in mediating peripheral metabolic adaptations during and after exercise training.

DISCUSSION

The mechanisms mediating long-lasting adaptations after exercise that enhance health and metabolism have been a focal point of physiology research. The conventional view is that following training, these processes take place in the periphery at skeletomuscular,^3–5 cardiovascular,^5,6 metabolic,^8–10 immune,^11,12 and endocrine-related sites.^13,14 Identification of molecules, like myokines,^50,51 released from peripheral tissues after exercise has led to the exploration of exercise mimetics^52,53 that achieve exercise-like effects in the absence of physical activity. The promise of these molecules strengthened the view that exercise-dependent endurance is driven by peripheral signals. Our results add to the understanding of how exercise leads to healthy adaptations by demonstrating that 1) VMH SF1 neurons are activated by exercise, 2) increased excitatory drive onto VMH SF1 neurons increases SF1 output as training progresses, and 3) VMH SF1 neuron activity is essential for the development of exercise-dependent endurance. This evidence demonstrates a significant role for the brain in enhancing exercise endurance and coordinating peripheral metabolic adaptations.

Just as exercise remodels muscle, we observe a parallel remodeling in the excitatory drive on VMH SF1 neurons that coordinates metabolic function and endurance-like adaptations in response to exercise. While prior studies have established that exercise enhances neurogenesis,^15–17 synaptic connectivity,^18–22 and overall cognitive function,^52,54 our work demonstrates that central neural adaptations contribute directly to metabolic and physiological adaptations following exercise. The current study supports a feed-forward model^55,56 in which VMH SF1 neurons activated by exercise undergo plasticity to potentiate responses to subsequent exercise training. Our electrophysiological and morphological studies suggest that exercise-induced plasticity results in increased excitability of SF1 neurons along with a higher frequency of EPSCs and a higher density of synapses on dendritic spines (Figures 3E, 3F, and 3H–J). Thus, our results identify VMH SF1 neurons as a structural and functional plasticity node that appears to store exercise history and drive the physiological adaptations underlying endurance. These changes further enable greater amounts of VMH SF1 output during subsequent exercise sessions.

The role of VMH SF1 neurons in exercise builds on prior work demonstrating the importance of the VMH in facilitating adaptations to metabolic demands^57–59 and regulating metabolism during exercise.^60–63 VMH SF1 neurons have an essential role in the maintenance of proper body weight, food intake, energy expenditure, adiposity, glycemia, and insulin sensitivity.^{33–35,37,64} Genetic perturbations demonstrate that proper functioning of SF1 neurons is important for exercise performance and training improvements.^37,65,66

Why are SF1 neurons a critical nexus for training improvements? Exercise acutely depletes glycogen stores in liver and muscle tissue. Increased glucose availability following exercise promotes glycogen synthesis, and consequentially, exercise recovery. Because SF1 neural stimulation increases blood glucose,⁴⁸ one potential consequence of elevated SF1 neural activity following exercise is to promote recovery. Furthermore, endurance capacity increases are initiated by a cascade of changes in skeletal muscle including the upregulation of PGC-1α which increases oxidative capacity, generates more efficient fuel utilization, and improves glucose handling.^67–69 Activity manipulations have shown a role for VMH SF1 neurons in driving muscle PGC-1α expression⁶⁸ and glucose homeostasis.⁴⁸ Here, we demonstrate that VMH SF1 neuron activity following exercise training is required for the underlying transcriptional changes in muscle that precede improvements in endurance capacity. Given the role of SF1 neural activation in increasing blood glucose and inducing PGC-1α, it is plausible that endogenous SF1 neuronal activity after exercise may generate the signals for recovery and subsequent adaptations that enhance future performance.⁷⁰

How do VMH SF1 neurons fit within a broader neural network that promotes physiological adaptations? VMH SF1 neurons have multiple receptor-defined subclasses^34,58,71,72 and widespread projections across the forebrain and hindbrain.^73,74 Our in vivo recording data suggest that neurons activated by exercise are distinct from the SF1 neurons activated in response to social interaction (Figure S3),⁵⁵ suggesting functional specificity embedded within the SF1 population. We find that exercise-activated SF1 neurons are activated either during or at the cessation of exercise (Figure 2B). Over multiple exercise sessions, the population that becomes active at exercise cessation expands at the expense of the population that is active during exercise (Figure 2I). However, the pattern of activation in individual neurons tracked between sessions is variable, suggesting a dynamic plasticity that recruits a larger population of SF1 neurons while maintaining physiological separability based on activity patterns. Future experiments should seek to identify the neuroanatomical or physiological mechanisms that activate distinct SF1 neurons during and following exercise. Our discovery that VMH SF1 neuron stimulation enhances adaptations to repeated exercise suggests a potential circuit-based strategy to identify the neural networks that promote the physiological benefits of repeated exercise. Equally important work will involve identifying the peripheral → central pathways by which SF1 neurons are activated following exercise and determining the circuits that activate the excitatory inputs onto SF1 neurons.⁷⁵ Ultimately, a comprehensive understanding of how SF1 neurons are activated by exercise and how they influence networks that control metabolism will provide a more complete picture of how circuits in the brain are involved in optimizing exercise performance.

By emphasizing the role of the central nervous system, particularly VMH SF1 neurons, this work suggests that central neural adaptations contribute more significantly to endurance than previously recognized. This line of investigation raises the intriguing possibility of developing exercise mimetics that activate VMH SF1 neurons or their downstream pathways, offering a novel strategy to enhance endurance and metabolic function in populations unable to engage in regular physical activity. Ultimately, the interplay between central and peripheral mechanisms could lead to more comprehensive approaches to improving physical endurance and overall health.

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to J. Nicholas Betley (jnbetley@sas.upenn.edu).

Materials Availability

This study did not generate any new unique reagents.

Data and Code Availability

All data used in the figures have been deposited at Open Science Framework and are publicly available as of the date of publication. The DOI is listed in the Key Resources Table. Raw data will be shared by the Lead Contact upon request. This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request.

KEY RESOURCE TABLE

Reagent or Resource	Source	Identifier
Antibodies
Anti-Homer1	Synaptic Systems	Cat#160 003
Goat Anti-Rabbit 555	Invitrogen	Cat#A-21428
Bacterial and virus strains
AAV1-hSyn-FLEX-TeLC-P2A-eYFP-WPRE	Fan Wang (unpublished); NeuroTools (viral prep)	Addgene plasmid #135391
AAV1-CAG-FLEX-eGFP	Ian Wickersham (unpublished)	Addgene AAV1; 59331-AAV1
AAV1-hSyn1-SIO-stGtACR2-FusionRed	Mahn et al.79	Addgene AAV1; 105677-AAV1
AAV1-hSyn-FLEX-GCaMP6s-WPRE-SV40	Chen et al.78	Addgene AAV1; 100845-AAV1
AAV5-EF1a-double floxed-hChR2(H134R)-eYFP-WPRE-HGHpA	Karl Deisseroth (unpublished)	Addgene AAV5; 20298-AAV5
AAV5-EF1a-DIO-eYFP	Karl Deisseroth (unpublished)	Addgene AAV5; 27056-AAV5
AAV1-CAG-DIO-mCherry-mBDNF-shRNAmir	Vector Biolabs	shAAV-253926
Chemicals, Peptides, and Recombinant Proteins
Isoflurane	Piramal	Cat#0010250P
Zetamine (ketamine), 100 mg/mL	Med-Vet	Cat#RXKETAMINE
Rumpun (xylazine), 100 mg/mL	Med-Vet	Cat#RXXYLAZINE-RUMP
Loxicom (meloxicam), 5 mg/mL	Norbrook	Cat#55529-040-11
RNAlater	Millipore Sigma	Cat#R0901
Chloral hydrate	Millipore Sigma	Cat#C8383
Sodium chloride	Fisher Chemical	Cat#S271-3
Potassium chloride	Millipore Sigma	Cat#P9541
Magnesium chloride	Millipore Sigma	Cat#M4880
Calcium chloride	Millipore Sigma	Cat#C5670
Sodium phosphate monobasic	Millipore Sigma	Cat#S3139
Sodium bicarbonate	Millipore Sigma	Cat#S8875
Glucose	Millipore Sigma	Cat#G7528
Alexa Fluor 350 hydrazide	ThermoFisher	Cat#A10439
Potassium gluconate	Millipore Sigma	Cat#G4500
HEPES	Millipore Sigma	Cat#54457
EGTA	Millipore Sigma	Cat#E3889
Magnesium ATP	Millipore Sigma	Cat#A9187
Paraformaldehyde	MP Biomedicals	Cat#150146
Agarose	Millipore Sigma	Cat#A9539
Polyvinyl alcohol mounting medium with DABCO	Millipore Sigma	Cat#10981
Deposited data
Data reported in the figures	Open Science Framework	https://osf.io/3879d/overview?view_only=398cc5b1545f4b07ab29ddf6ceba1fbc
Experimental Models: Organisms/Strains
Mouse: Tg(Nr5a1-Cre)7Lowl/J	The Jackson Laboratory	RRID: IMSR_JAX:012462
Mouse: Ai9 B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J	The Jackson Laboratory	RRID: IMSR_JAX:007909
Mouse: SF1-eGFP	Stallings et al. (2002)77; Teppei Fujikawa, Keith L. Parker	N/A
Mouse: C57BL6/J	The Jackson Laboratory	RRID: IMSR_JAX:000664
Oligonucleotides
Mm-Nr5a1 (SF1) RNAscope probe	ACDbio	Cat#445731-C2
Mm-Bdnf RNAscope probe	ACDbio	Cat#424821
Software and Algorithms
Matlab R2024a	MathWorks	https://www.mathworks.com/products/matlab
RStudio 1.2.5019	Posit	https://posit.co/download/rstudio-desktop/
Bioconductor 3.8	Bioconductor	https://bioconductor.org/
CellProfiler	Stirling et al. (2021)83	https://cellprofiler.org/
MetaScreen	Sable Systems	https://www.sablesys.com/products/promethion-high-definition-room-calorimetry-system/promethion-software/
MacroInterpreter	Sable Systems	https://www.sablesys.com/products/promethion-high-definition-room-calorimetry-system/promethion-software/
Synapse	Tucker-Davis Technologies	https://www.tdt.com/component/synapse-software
Prism 10	GraphPad	https://www.graphpad/com
Minian	Dong et al. (2022)94	https://github.com/denisecailab/minian
CellReg	Sheintuch et al. (2017)42	https://github.com/zivlab/CellReg
pCLAMP 11	Molecular Devices	https://www.moleculardevices.com/products/axon-patch-clamp-system/acquisition-and-analysis-software/pclamp-software-suite
Easy Electrophysiology	Easy Electrophysiology	https://www.easyelectrophysiology.com
Other
Vanderbilt Mouse Metabolic Phenotyping Center	Vanderbilt University	RRID: SCR_021939
Purina 5001	Lab Diet	N/A
RNAscope manual assay	ACDbio	N/A
TruSeq stranded mRNA kit	Illumina	Cat#20020595
TruSeq Unique dual indexes	Illumina	Cat#20022371
NextSeq 2000	Illumina	N/A
4200 TapeStation	Agilent	Cat#G2991BA
Qubit 4 Fluorometer	Invitrogen	Cat#Q33226
Dual Just for Mouse Stereotax	Stoelting	Cat#51733
Syringe pump for stereotaxic injections	Harvard Apparatus	Cat#703007 PHD Ultra
Nanofil syringe	World Precision Instruments	Cat#NAN0FIL-100
Metabond	Parkell	Cat#S380
Ortho-Jet	Lang Dental Manufacturing	Cat#1323
DEXA InAlyzer2 System	Micro Photonics	N/A
Optic fiber for optogenetics	Thorlabs	Cat#FT200UMT
Patch cable for optogenetics	Thorlabs	Cat#M72L02
Ferrule for optogenetics	Kientec Systems	Cat#FZI-LC-230
Optic fibers for photometry	Doric	Cat#MF2.5, 400/430-0.37
GRIN lens, 8mm, 600 μm diameter	Inscopix	Cat#1050-004600
KwikSil	World Precision Instruments	Cat#KWIK-SIL
Miniscope baseplate v4	Open Ephys	Cat#OEPS-7416
Promethion	Sable Systems	N/A
Minispec Body Composition Analyzer	Bruker	Cat#LF50N/A
Phenomaster	TSE Systems	N/A
Mating sleeve for optogenetics	Kientec Systems	Cat#SZI-LC-SP LC
473nm laser	Lasever	Cat#LSR473H
Optical power meter	Thorlabs	Cat#PMD130D
Motorized treadmill	Columbus Instruments	Cat#Exer-6
Flying Saucer Running Wheel (small)	Ware Pet Products	Cat#03281
Integrated fiber photometry system	Tucker-Davis Technologies	Cat#RZ10x
Connector for fiber photometry	Thorlabs	Cat#ADAF2
USB camera	ELP	Cat#ELP-USB100W05MT-DL36
Miniscope camera v4.4	Open Ephys	Cat#OEPS-7407
Miniscope DAQ v3.3	Open Ephys	Cat#OEPS-7431
Miniscope dummy scope, v4	Open Ephys	Cat#OEPS-7409
Vibratome	Leica Biosystems	Cat#VT1000S
Axioskop FS2	Zeiss	N/A
QuantEM:512SC camera	Teledyne Photometrics	N/A
Axopatch 700B amplifier	Molecular Devices	N/A
DM6	Lecia Microsystems	N/A
Ultra 2 glucose meter	OneTouch	N/A
Ultra glucose test strips	OneTouch	N/A

STAR Methods

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

Adult (at least 6 weeks old) male and female mice were group-housed on a 12 h light/12 h dark cycle in a temperature-controlled environment with ad libitum access to food (Purina 5001) and water. SF1-Cre (Jackson Laboratory stock # 012462, Tg(Nr5a1-Cre)7^Lowl/J)³⁴ and C57BL6/J mice were used for metabolic, physiological, and exercise experiments. For whole-cell patch clamp recordings, SF1-Cre mice were crossed with Ai9 (Jackson Laboratory stock # 007909, B6.Cg-Gt(ROSA)26Sor^{tm9(CAG-tdTomato)Hze}/J)⁷⁶ mice to drive tdTomato expression in SF1 neurons. SF1-eGFP mice⁷⁷ were also used for whole-cell patch clamp recordings. Only male mice were used in whole-cell patch clamp and endoscopic recordings. Both male and female mice were used for all other experiments, and a roughly equal number of males and females were assigned to experimental conditions. Mice were otherwise randomly assigned to experimental conditions unless noted. In experiments with male and female mice, results were not the effect of sex. All procedures were approved by the Institutional Animal Care and Use Committees of the University of Pennsylvania, the University of Texas Southwestern Medical Center, The Jackson Laboratory, and Vanderbilt University.

METHOD DETAILS

Recombinant Adeno-Associated Virus (rAAV)

The following rAAVs were used for these experiments: AAV1-hSyn-FLEX-TeLC-P2A-eYFP-WPRE (titer = 3.6×10¹³ GC/mL, Addgene 135391 packaged by Neurotools) was used for TetTox Light Chain neural silencing experiments; AAV1-CAG-FLEX-eGFP (titer = 1.0×10¹³ GC/mL, Addgene 59331) was used as the control. AAV1-CAG-DIO-mCherry-mBDNF-shRNAmir (titer = 8.8 × 10¹² GC/mL, Vector Biolabs shAAV-253926) was used for Bdnf knockdown experiments. AAV1-hSyn-FLEX-GCaMP6s-WPRE-SV40 (titer = 2.6×10¹³ GC/mL, Addgene 100845) was used for fiber photometry recording of VMH SF1 neurons and for synapse reconstruction experiments. Diluted (1:3) AAV1-hSyn-FLEX-GCaMP6s-WPRE-SV40 was used for endoscope recordings of VMH SF1 neurons. AAV1-hSyn1-SIO-stGtACR2-FusionRed (titer = 2 ×10¹³ GC/mL, Addgene 105677) was used for optogenetic inhibition of VMH SF1 neurons. AAV5-EF1a-double floxed-hChR2(H134R)-eYFP-WPRE-HGHpA (titer = 2.5×10¹³ GC/mL, Addgene 20298) was used for optogenetic stimulation of VMH SF1 neurons. AAV-EF1a-DIO-eYFP (titer = 3.5×10¹² GC/mL, Addgene 27056) was used for fluorescent protein-only controls in optogenetics experiments.

pAAV-hSyn-FLEX-TeLC-P2A-eYFP-WPRE was a gift from Fan Wang (Addgene plasmid 135391). AAV1-CAG-FLEX-eGFP was a gift from Ian Wickersham (Addgene viral prep # 59331-AAV1). AAV1-hSyn-FLEX-GCaMP6s-WPRE-SV40⁷⁸ was a gift from Douglas Kim & GENIE Project (Addgene viral prep # 100845-AAV1). AAV1-hSyn1-SIO-stGtACR2-FusionRed⁷⁹ was a gift from Ofer Yizhar (Addgene viral prep # 105677-AAV1). AAV5-EF1a-double floxed-hChR2(H134R)-eYFP-WPRE-HGHpA was a gift from Karl Deisseroth (Addgene viral prep # 20298-AAV5). AAV5-Ef1a-DIO EYFP was a gift from Karl Deisseroth (Addgene viral prep # 27056-AAV5).

Stereotaxic Surgery

Stereotaxic surgeries were performed as previously described.^80–82 For all surgical procedures, mice were anesthetized with isoflurane (3.0–4.0%), given s.c. meloxicam (5 mg/kg, UPenn) or carprofen/buprenorphine (5 mg/kg and 3.25 mg/kg, JAX) for analgesia, had their scalp hair removed, and were placed into a stereotaxic device. Isoflurane was administered at 1.0–3.0% throughout surgery to maintain a constant surgical plane. A midline incision was performed to expose the skull, and a craniotomy was drilled above the injection and implant site, if applicable. All coordinates are given relative to bregma, and injections were performed using a Nanofil syringe (World Precision Instruments) or with glass pipette back filled with mineral oil and connected to a pressure injector (Harvard Apparatus or Drummond Nanoject II). The maximum injection rate for viral suspension was 100 nL/min. All mice were given a minimum of 3 weeks for recovery and viral expression before beginning experiments. Injection sites, fiber, and lens placements were verified post-mortem using fluorescence microscopy (see below).

SF1 TetTox neural silencing and SF1 Bdnf knockdown experiments

A glass pipette back filled with mineral oil and connected to an injector (Harvard Apparatus) was used to pressure-inject 400 nL per hemisphere of an rAAV encoding Cre-dependent tetanus toxin light chain (TetTox) or a total of 300 nL per hemisphere of an rAAV encoding Cre-dependent mBDNF shRNA bilaterally into the VMH (−1.50 AP; 0.25 ML; −5.65, −5.50 DV; 200 nL/150 nL per site, respectively) of SF1-Cre mice and Cre-negative littermates as a control.

Fiber photometry experiments

A glass pipette back filled with mineral oil and connected to an injector (Harvard Apparatus) was used to pressure-inject 400 nL of a rAAV encoding Cre-dependent GCaMP6s unilaterally into the VMH (−1.50 AP; 0.25 ML; −5.65, −5.50 DV; 200 nL per site) of SF1-Cre mice. In the same surgery, a 400 μm, 0.37 NA fiber optic (Thorlabs) connected to a 2.5 mm diameter stainless steel ferrule (Doric) was implanted 0.25 mm above the more dorsal injection site.

Endoscopic imaging experiments

Surgical procedures began with an injection of an rAAV encoding Cre-dependent GCaMP6s diluted 1:3 in saline, injected unilaterally into the VMH (−1.50 AP; 0.25 ML; −5.65, −5.50 DV; 150 nL per site) of SF1-Cre mice, as described above. After allowing 3 weeks for expression, a gradient index (GRIN) lens implantation was performed. The VMH craniotomy was widened, and a blunted 22 G needle was lowered slowly (0.2 mm/min) to 0.25 mm dorsal to the most dorsal injection site, left still for 10 min, and then removed at the same rate. A GRIN lens 8.4 mm in length and 500 μm in diameter (Inscopix) was slowly lowered to approximately 0.15 mm above the most dorsal injection site. The final depth was determined by visualizing GCaMP-expressing neurons through the miniature microscope (v4.4, Open Ephys) positioned over the lens. The lens and head bar were secured with Metabond (Parkell) and black dental acrylic (Lang Dental Manufacturing), leaving only the top surface of the lens exposed. Parafilm secured with KwikSil (World Precision Instruments) was placed over the top of the lens for protection between surgeries. After allowing at least 3 weeks for recovery, a baseplate (v4, Open Ephys) was secured over the lens with black dental acrylic. The plane for baseplate fixation was determined by visualizing GCaMP-expressing neurons through the miniature microscope (v4.4, Open Ephys) attached to the baseplate.

Synapse reconstruction experiments

A glass pipette back filled with mineral oil and connected to an injector (Harvard Apparatus) was used to pressure-inject 400 nL of a rAAV encoding Cre-dependent GCaMP6s unilaterally into the VMH (−1.50 AP; 0.25 ML; −5.65, −5.50 DV; 200 nL per site) of SF1-Cre mice.

Optogenetic SF1 inhibition experiments

A glass pipette back filled with mineral oil and connected to an injector (Harvard Apparatus) was used to pressure-inject 500 nL per hemisphere of an rAAV encoding a Cre-dependent soma targeted anion-conducting ChR2 (stGtACR2) at a 15° angle (−1.50 AP; +/−1.90 ML; −6.0, −5.8 DV; 250 nL per site) of SF1-Cre mice. Fibers (200 μm, 0.37 NA, Thorlabs) coupled to 1.25 mm zirconium ferrules (Kientec Systems Inc.) were implanted bilaterally at a 15° angle targeted above the VMH injection sites (−1.50 AP; +/−1.85 ML; −5.70 DV). For both optogenetic and photometry fiber implants, the skull surface was scored, and fiber optics were secured in place with a layer each of Metabond (Parkell) and dental acrylic (Lang Dental Manufacturing).

Optogenetic SF1 stimulation experiments

A Nanofil syringe (World Precision Instruments) was used to pressure-inject 500 nL of an rAAV encoding Cre-dependent ChR2 unilaterally into the VMH (−1.50 AP; 0.25 ML; −5.65, −5.50 DV; 250 nL per site) of SF1-Cre mice. 500 nL of a fluorophore-only rAAV was similarly injected into SF1-Cre mice to be used as a control population. A 200 μm, 0.37 NA fiber (Thorlabs) coupled to a 1.25 mm zirconium ferrule (Kientec Systems Inc.) was implanted unilaterally above the VMH injection site (−1.50 AP; 0.25 ML; −5.45 DV). The implants were secured as described above for optogenetic inhibition experiments.

Post-hoc Histology to Assess Viral Expression, Fiber, and Lens Placement

For histological verification of viral expression and implants, mice were transcardially perfused with 10 mL PBS and 10 mL 4% paraformaldehyde (PFA). Brains were removed and post-fixed in 4% PFA overnight. Brains were then washed in PBS, set in 4% agarose, and sectioned into 100 μm on a vibratome (VT1000s, Leica Biosystems). Sections were mounted to slides in polyvinyl alcohol medium with DABCO (Millipore Sigma) and imaged using an epifluorescence microscope (DM6, Leica Microsystems) to confirm viral expression, fiber, and lens placement.

Exercise Protocols

Treadmill exercise

For all treadmill exercise experiments except the indirect calorimetry exercise stress test (Figure 1 and S1; described in ‘VMH SF1 TetTox Exercise Experiments: Stress Test’ section below), a 6-lane motorized treadmill (Exer-6; Columbus Instruments) was used. For all treadmill exercise experiments except the indirect calorimetry stress test and electrophysiology experiments, mice were coaxed to remain running on the treadmill with gentle prodding. For the indirect calorimetry stress test and electrophysiology experiments, an electrified grid was activated at the end of the treadmill to administer an electrical shock (1.0 mA) if the animal exited the treadmill. In all treadmill experiments, running was constantly monitored by an experimenter. Sedentary mice were exposed to the fixed treadmill (i.e., moving at 0 m/min). All exercised mice were first acclimated to the treadmill using a 3-day protocol prior to starting experiments: 3 h at 0 m/min (Day 1); 2 h at 0 m/min, 15 min at 6 m/min, 5 min at 8 m/min, 5 min at 10 m/min (Day 2); 1 h at 0 m/min, 15 min at 10 m/min, 5 min at 8 m/min, 15 min at 10 m/min (Day 3). The remainder of the protocol is detailed in Tables S1 and S2; all protocols were run with 0° incline. To assess endurance capacity and the physiological responses to exercise, mice were run on endurance tests in which running speed gradually increased (Table S1) until mice reached exhaustion as described previously.³⁷ Exhaustion was operationally defined as the point at which mice remained immobile against the end of the treadmill for 3 s without forward locomotion despite prodding by the experimenter.

Voluntary exercise

In voluntary running experiments, all control (sedentary) mice had free access to fixed wheels in their home cage, while exercise mice had free access to live wheels. Mice in these experiments were singly housed to ensure accurate data collection of distance run for each hour.

In Situ Hybridization

Mice were euthanized 30 min after being removed from the treadmill by transcardial perfusion first with 10 mL PBS followed by 10 mL 10% neutral buffered formalin (NBF). Brains were fixed in 10% NBF for 24 h at room temperature, paraffin embedded and sectioned on a microtome at 12 μm. RNAscope was performed using probes for Bdnf (channel 1) and Nr5a1 (SF1, channel 2) and stained with DAPI according to manufacturer (ACDbio) instructions. Sections were imaged at 63x (oil immersion) objective on an epifluorescence microscope (DM6, Leica Microsystems). Exposure settings were kept constant between sections. Images were cropped in ImageJ to include only VMH (boundaries determined by Allen Brain Atlas) and further segmented into dorsomedial/central and ventrolateral subregions for quantification. Analysis was performed in CellProfiler⁸³ using a custom quantification pipeline⁸⁴ that quantified the number of DAPI cells and the number of dots (transcripts) for each channel within each DAPI-labeled cell. DAPI cells and RNA transcripts were both detected based on intensity and pixel thresholds, determined and validated manually. A cell was considered Sf1+ or Bdnf+ if it contained 2 or more Sf1 or Bdnf transcripts, respectively.

VMH SF1 TetTox Exercise Experiments

Stress test with indirect calorimetry

Individual TetTox expressing mice (SF1-Cre+) and littermate controls (SF1-Cre-negative) were placed on a treadmill inside of a metabolic chamber to perform indirect calorimetry during an exercise stress test (Sable Systems). The treadmill was at no incline (0°) for the duration of the stress test. Following a 20 min baseline period, the treadmill began moving at 10 m/min. The speed was increased by 4 m/min every 3 min until the animals reached exhaustion. Exhaustion was defined as mice remaining on the shockpad for > 5 s. Upon exhaustion, the treadmill was returned to a standstill and mice remained in the chamber for 15 min of recovery. Air was sampled through microperforated stainless steel sampling tubes that ensure uniform cage air sampling with an excurrent flow rate of 3000 L/min. Gases were continuously measured and averaged in 15 s intervals.

Muscle tissue RNAseq

TetTox-expressing (SF1-Cre+) mice and littermate (Cre-negative) controls completed the 3-week exercise regimen as described in Tables S1 and S3. During training, sedentary controls were placed on the treadmill at standstill. On week 4, mice repeated week 3 training to maintain fitness levels (Table S3). On day 5 of week 4, muscle was harvested after mice ran to 90% exhaustion (based on week 3 endurance) to minimize physiological effects of overexertion. Three hours after the end of this final endurance session—a time window previously shown to capture exercise-induced gene expression changes in muscle⁴¹—mice were anesthetized with a ketamine/xylazine cocktail. Immediately after confirmation of anesthesia, the entire quadricep of one leg was extracted. After removing any tendons, fascia, and fat stuck to the tissue, it was rinsed with cold sterile PBS to remove excess blood and then placed immediately into RNAlater. Tissue was stored at −80°C for further use. Tissue was collected from exercised and sedentary cage mate controls in an alternating fashion.

Libraries were prepared using the Illumina TruSeq stranded mRNA kit with IDT for Illumina TruSeq Unique dual indexes according to the manufacturer’s instructions. Quality and quantity control of RNA and libraries were performed using Agilent 4200 TapeStation and Qubit 4, respectively. Libraries were sequenced on an Illumina NextSeq 2000 to produce 75–base pair single end reads.

Exercise training and endurance tests

Endurance tests were performed as described in Table S1 and were run before training commenced and on the sixth day during each week of training. In between tests, mice were trained on a 5 day per week regimen for 3 weeks. The duration of runs increased from 20 to 30 min over the first week and were 60 min on all subsequent weeks. Running speeds gradually increased over the course of each exercise session. The training protocol intensity (i.e., starting/ending speed) was lower relative to other experiments to minimize any potential training gap between TetTox and control animals. The detailed exercise protocol is provided in Table S2.³⁷ While most TetTox mice could complete the initial training sessions, 2 minute breaks in training were allowed as the training load increased as some individuals were unable to complete the full training session. Littermate controls were also given the same breaks in training.

SF1 Bdnf knockdown

SF1-Cre mice expressing the Bdnf knockdown (KD) vector and Cre-negative littermate controls underwent endurance tests (Table S1) and exercise training (Table S2). Bdnf KD mice and their Cre-negative littermates could complete all training sessions with minimal breaks. To quantify body fat percentage, dual-energy X-ray absorptiometry (DEXA) scans (DEXA InAlyzer2 System, Micro Photonics Inc.) were performed at baseline, during which SF1-Cre mice expressing the Bdnf knockdown (KD) vector and Cre-negative littermate controls were anesthetized with isoflurane (3.0–4.0%) and placed on the scan platform. Body position was aligned to a reference. Mice remained under anesthesia for the entirety of the scan. Posthoc in situ hybridization was used to validate the knockdown (see above).

Calcium Imaging

Fiber photometry

Dual wavelength fiber photometry was performed as previously described.²⁰ Here, we recorded SF1 neuron GCaMP signal before, during, and after exercise. 470 and 405 nm light, which excite calcium-dependent and -independent fluorescence respectively, were generated through fiber-coupled LEDs (LUX, Tucker Davis Technologies) and modulated by an integrated processor (LUX RZ10X, Tucker Davis Technologies) at 211 and 566 Hz, respectively. The combined excitation light was delivered through a 400 μm, 0.48 NA optical fiber (Doric), which was securely connected to the implanted fiber (ADAF2, Thorlabs). GCaMP emissions fluorescence was collected through the same patch cord and focused onto a LUX photosensor in the same integrated processor. Emission signals were converted to electrical signals, sampled at 1,000 Hz, and demodulated by the RZ10x integrated processor. 1-megapixel USB cameras (ELP-USB100W05MT-DL36, ELP) were used to record mouse behavior during recording. GCaMP fluorescence signals were set to similar levels across mice by adjusting the 470 and 405 nm output power. During photometry recordings, the exercise training protocols used in other experiments were modified slightly. Mice performed a 5 d per week, 3-week long training schedule in which the duration of runs increased from the first to second and third weeks, and speed increased within each individual run (Table S3). Endurance tests were run prior to training and again following a two-day rest period after week 3.

Endoscopic calcium imaging

For endoscopic calcium imaging (miniscope), mice were habituated to a dummy miniscope 2 hours per day for 1 week prior to the first endurance test. On days 1–2 of habituation, a dummy miniscope was attached to the baseplate, and mice were placed back in their home cages with the scope for 2 hours. On days 3–5 of habituation, the dummy miniscope was attached, and mice were placed in the treadmill at standstill for 2 hours. On days 6–7, the dummy miniscope was attached, and mice were placed in the treadmill and completed 5, 2 min slow walks (4 m/min) with the scope, spaced over 30 min. Mice completed the rest of the 2-hour habituation period in the treadmill at standstill.

Prior to recording sessions, the miniscope was attached and mice were placed on the treadmill at standstill for 30 min relative to the start of baseline recording to re-acclimate. Imaging data was acquired as previously described.⁸⁵ The software associated with Miniscope data acquisition (DAQ) unit (v3.3, Open Ephys) was used to control illumination, focus, and collect data. 470 nm light was generated from an LED integrated within the miniscope and shone through the GRIN lens. Images were acquired at 20 fps. LED power was optimized for each mouse to maximize the dynamic range of pixels in putative neurons, and this power was kept consistent across recordings for each mouse. Mouse behavior was monitored and acquired at 50 fps using a 4-megapixel infrared camera (MiniCAM, Open Ephys) connected to a second DAQ integrated with the same software. Imaging sessions were performed while the mouse ran on the treadmill.

The exercise training protocol was modified for endoscopic imaging. After a baseline endurance recording session (Table S1), mice were given two days to rest. Because the weight of the baseplate and miniscope dramatically reduced endurance (approx. 2 g, which represents ~10% of the animal’s body weight), training intensity was reduced, allowing mice to complete the training protocol. On the first day of training (Day 4), mice ran 5 min at 6 m/min, 2 min 7 m/min, 2 min at 8 m/min, 2 min at 9 m/min, 2 min at 10 m/min, and 47 min at 11 m/min. As mice were individually trained, they were given breaks (2 min) as needed. If three breaks occurred within the span of 15 min, the treadmill speed was lowered by 1 m/min. Because endoscopic animals were each trained individually, the training protocol was personalized for each animal’s ability. The distance at the end of the hour was used as the baseline training distance. On the next two training days (Days 5–6), distance was increased by 5–10% per day depending on individual ability/animal reaching exhaustion. On the third training day (Day 7), there was no increase in distance to ensure mice were adequately recovered before the second endurance test (Day 8).

To examine neural activity from the same neurons in response to social interaction, mice were acclimated to the apparatus and miniscope for 30 min. After a baseline recording, an unfamiliar conspecific of the same sex and approximately the same age was placed into the opposite end of the apparatus facing the test mouse for one minute. The conspecific was previously habituated to the testing apparatus.

Whole-cell Patch Clamp Electrophysiology

Male SF1-Cre::Ai9 mice or SF1-eGFP mice⁷⁷ aged 8–16 weeks were used. Body weight-matched animals were assigned to exercise and sedentary groups.

In treadmill training experiments, mice assigned to the exercise group completed the 3-week long training paradigm as used in previous experiments (Table S3), with the exception that mice received a 1.0 mA foot shock if they ceased running. Sedentary mice remained on the treadmill for 1 h at 0 m/min and received foot shock if they stepped off the treadmill. Immediately following the final endurance run (Table S1), mice were sacrificed and brain slices were prepared for recording. In voluntary running patch clamp experiments, the body composition of each mouse was recorded before introduction of the wheel using a Bruker minispec LF50 body composition analyzer. All wheels, active or disabled, were left within the animals’ standard home cage for 4 weeks. The day prior to sacrificing the animal for electrophysiological recordings, a final body composition measurement was taken.

For all whole-cell patch clamp recordings, acute brain slices brain sections were prepared for recording as previously described.^{20,21,86–90} Briefly, mice were deeply anesthetized with i.p. injection of 7% chloral hydrate immediately following exercise and transcardially perfused with a modified ice-cold artificial cerebrospinal fluid (ACSF; 126 mM NaCl, 2.8 mM KCl, 1.2 mM MgCl₂, 2.5 mM CaCl₂, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃, and 5 mM glucose). The mice were then decapitated, and the entire brain was removed and immediately submerged in ice-cold, carbogen-saturated (95% O₂ and 5% CO₂) ACSF. Coronal sections (250 μm) were cut with a vibratome (Leica VT1000S) and then incubated in oxygenated ACSF (32 °C-34 °C) for at least 1 h before recording. The slices were bathed in oxygenated ACSF (32 °C-34 °C) at a flow rate of ~2 mL/min. All electrophysiology recordings were performed at room temperature.

The internal solution included 120 mM K-gluconate, 10 mM KCl, 10 mM HEPES, 5 mM EGTA, 1 mM CaCl₂, 1 mM MgCl₂, and 2 mM MgATP, and 0.03 mM Alexa Fluor 350 hydrazide dye (pH 7.3). Epifluorescence was briefly used to target fluorescent cells, at which time the light source was switched to infrared differential interference contrast imaging to obtain the whole-cell recording (Zeiss Axioskop FS2 Plus equipped with a fixed stage and a QuantEM:512SC electron-multiplying charge-coupled device camera). Electrophysiological signals were recorded using an Axopatch 700B amplifier (Molecular Devices), low-pass filtered at 2–5 kHz, and analyzed offline (pCLAMP, Molecular Devices). Membrane potentials and firing rates were measured from SF1 neurons in brain slices. Recording electrodes had resistances of 2.5–5.0 MΩ when filled with the K-gluconate internal solution. Neurons were voltage-clamped at −70 mV (for excitatory postsynaptic currents) and −15 mV (for inhibitory postsynaptic currents). Frequency and peak amplitude of excitatory and inhibitory currents were analyzed by using the Easy Electrophysiology program (Easy Electrophysiology Ltd). Membrane potential values were not compensated to account for junction potential (−8 mV).

Synapse Reconstruction

Mice were transcardially perfused with 4% paraformaldehyde (PFA), and brains were removed and postfixed for 36 h in cold 4% PFA then stored in 0.1 M PB. 50 μm-thick slices were made on a vibratome, incubated with anti-Homer 1 (Rabbit anti-Homer1, Synaptic Systems, diluted 1:2000 in blocking buffer) followed by fluorescent secondary antibodies (Goat anti-Rabbit 555, Invitrogen, diluted 1:400 in blocking buffer), then mounted and coverslipped. Images were acquired on a Leica SP8 confocal microscope equipped with Lightning deconvolution and a 63x, 1.4 NA objective. Z-stacks were collected with 50 nm x-y pixel sizes and with 0.2 μm steps. Dendrites and dendritic spines were reconstructed in NeuronStudio⁹¹ and Homer1 spine localization was analyzed in Fiji. All experimenters were blind to the experimental conditions during the analysis.

Optogenetic Inhibition

Photoinhibition was performed as previously described.⁹² The output beam from a 473 nm diode laser (Lasever) was manually controlled and applied continuously for 15 min per session. The laser was coupled to a 200 μm, 0.39 NA bifurcated multimode optical fiber (Thorlabs) using 1.25 mm OD zirconium mating sleeve (Kientec), allowing bilateral delivery of light into the brain. Laser power was measured with an optic power meter (Thorlabs) each day and set to 10 mW/ferrule for an irradiance of 15–50 mW/mm² from the fiber tip.

Exercise training and endurance tests

SF1-Cre mice and Cre-negative littermate controls injected with stGtACR2 underwent endurance tests (Table S1) and exercise training (Table S3). A baseline endurance test was run before training began and on the sixth day of each of the three training weeks. An endurance test was run before the start of training and on the sixth day of each of the three training weeks. Mice did not receive inhibition and were not tethered to fibers for endurance tests. For training sessions, both Cre+ and Cre-negative mice were tethered at the start of the session and received inhibition immediately at the end of the session, for 15 minutes.

Blood glucose measurements during/after exercise

On week 3 of training, blood glucose was measured from tail blood of trained Cre+ and Cre-negative mice before and after an exercise training session with post-exercise inhibition. Because blood glucose could only be measured from one mouse at a time and mice shared a treadmill, there was a measurement time lag of 15–60s post-exercise, split evenly between Cre+ and Cre-negative mice.

Optogenetic Simulation

Photostimulation of ChR2 expressing SF1 neurons was performed as previously described^68,93 with 10 ms pulses given at 20 Hz for 2 s, repeated every 4 s—a pattern that has been shown to elicit immediate early gene expression in SF1 neurons.⁶⁸ The output beam from a 473 nm diode laser (Lasever) was controlled by a microcontroller (Uno, Arduino). The laser was coupled to a 200 μm, 0.39 NA multimode optical fiber (Thorlabs) using a 1.25 mm OD zirconium mating sleeve (Kientec), allowing the delivery of light into the brain. Laser power was measured with an optic power meter (Thorlabs) each day and set to 20 mW for an irradiance of 25–100 mW/mm² from the fiber tip, as we have previously used to activate hypothalamic neurons.⁹³

Exercise training and endurance tests

SF1-Cre mice expressing ChR2 or eYFP (controls) underwent endurance tests (Table S1) and exercise training (Table S3). An endurance test was run before the start of training and on the sixth day of each of the three training weeks. Mice did not receive stimulation and were not tethered to fibers for endurance tests. For training sessions, both ChR2 and eYFP mice were tethered at the start of the session and received stimulation immediately at the end of the session, for 1 hour.

Blood glucose measurements during/after exercise

Blood glucose was measured from tail blood of trained ChR2 and eYFP mice before and after an exercise training session with stimulation. Because blood glucose could only be measured from one mouse at a time and mice shared a treadmill, there was a measurement time lag of 15–60s post-exercise, split evenly between ChR2 and eYFP mice.

Indirect calorimetry with optogenetic stimulation

Indirect calorimetry in untrained SF1 ChR2 mice was performed using PhenoMaster metabolic cages (TSE Systems). Mice were first acclimated to single housing in the recording cages for 7 days. The cages were then connected to the PhenoMaster system, where O₂ consumption, CO₂ production, and activity were continuously monitored. Gas exchange was sampled for one minute, every 12 min. Mouse activity detected by peripheral infrared sensors was averaged into matched 12 min bins. Mice underwent 2 stimulation trials and 2 sham (no light) trials each. Only one trial was performed per day. Trials were performed on alternating days at the same time during the light cycle. For each trial, 2 hours prior to stimulation/sham stimulation, the chamber lids were briefly opened, and mice were tethered to a fiber optic as described above. Food and water were removed. The fiber optic was fed through a small puncture in Parafilm in the chamber lid to minimize gas leak. In stimulation trials, mice received optogenetic stimulation for 1 hour as described above. The post-stimulation period was 1 hour. During this time, stimulation was turned off but mice remained tethered and without food/water. After the post-stimulation period, mice were detached, and ad libitum food/water was returned to the cages. Sham trials were identical to stimulation trials, but lasers remained off.

QUANTIFICATION AND STATISTICAL ANALYSIS

General Analysis

Data are expressed as mean ± SEM in all figures and text. Prism 10 (GraphPad) was used to calculate all statistics. Paired or unpaired t-tests were used for comparison of means in within- and between-subjects designs, respectively. A Mann-Whitney test was used for in situ hybridization comparisons. A Kolmogorov-Smirnov test was used for dendritic morphology comparisons. A linear model with Benjamini-Hochberg correction was used for RNA sequencing analysis. A simple linear regression was used to compare exercise performance with body fat percentage. Repeated measures ANOVAs were used to make comparisons across two groups or for repeated measures with the same group. Tukey and Sidak post-hoc comparisons were used when appropriate. Error bars and shading represent SEM. In all figures, ☼ p < 0.05, ☼☼ p < 0.01, ☼☼☼ p < 0.001, for main effects of ANOVA and mixed effects analyses; † p < 0.05, †† p < 0.01, ††† p < 0.001, for interaction effects in ANOVA and mixed effects analyses; * p < 0.05, ** p < 0.01, and *** p < 0.001 for post-hoc, Mann-Whitney, Kolmogorov-Smirnov, and t-tests. Effects with p < 0.05 were considered significant. Statistical tests, n, and p-values for each panel are listed in Table S4.

TetTox Muscle Tissue RNAseq

Raw reads were mapped to the mouse reference transcriptome (Ensembl; Mus musculus version 67) using Kallisto version 0.46.0. Subsequent analysis was carried out using the statistical computing environment R version 3.6.1 in RStudio version 1.2.5019 and Bioconductor version 3.8. Briefly, transcript quantification data were summarized to genes using the tximport package and normalized using the trimmed mean of M values (TMM) method in edgeR. Genes with < 1 CPM in n+1 of the samples, where n is the size of the smallest group of replicates, were filtered out. Differentially expressed genes were identified with linear modelling using limma (FDR ≤ 0.05; absolute logFC ≥ 1) after correcting for multiple testing using Benjamini-Hochberg.

Calcium Imaging

Fiber photometry

Data were exported from Synapse to MATLAB (MathWorks) using a script provided by Tucker-Davis Technologies. Custom MATLAB scripts were used to downsample the data to 250 Hz and to normalize 470 nm-dependent fluorescence to Ca²⁺ independent 405-nm dependent fluorescence to control for bleaching and motion artifacts. ΔF/F was calculated by dividing the instantaneous 470 nm-dependent fluorescence by the session mean 470 nm-dependent fluorescence. ΔF/F is expressed as a percentage, where 0% is the recording session mean fluorescence. To analyze the neural response to the end of exercise, data were normalized to the mean fluorescence −5 to −4 min relative to the end of exercise.

Endoscopic imaging (miniscope)

Endoscopic imaging videos were processed with Minian, an open-source Python package for miniscope analysis.⁹⁴ Through Minian, endoscopic calcium imaging data were denoised and motion corrected. Neuronal footprints and calcium transients were extracted using constrained nonnegative matrix factorization (CNMF). Minian parameters were kept consistent across all sessions for each mouse. Each cell was manually validated. CNMF-defined cells were excluded if they met one or more of the following: a) spatial overlap and temporal activity identical with another cell, b) spatial overlap and temporal activity entirely opposite with another cell, c) not spherically shaped, and d) low signal-to-noise transient. CNMF calcium transients were converted into MATLAB-readable data structures for further analysis.

All CNMF calcium transients were z-scored. Timestamps to the nearest frame for the start and end of exercise were recorded in the miniscope software in real time and then manually validated post-hoc from behavior. CNMF calcium transients were aligned to these timestamps.

Activity 5 min before and 5 min after the run stop were averaged. Change in average activity between these two periods (post-run activity - run activity) for each cell was thresholded. Our initial inspection of the raw data suggested that there was a robust increase in the number of neurons activated after exercise as animals became more fit/trained. Because this was a property we observed even in neurons that were marginally activated in untrained mice, we set a low threshold of 0.05z so that we could track changes in these neurons across training. Cells above a 0.05 z post-run increase were classified as “activated” and cells below a 0.05 z decrease were classified as “inhibited.” Cells with values in between this range (−0.05 z to 0.05 z) were classified as “no change.” A higher threshold (+/−0.2 z) was used for social interaction analysis. Social preference index was calculated as previously described.⁵⁵

Tracking between days: For each mouse, cells that could be tracked between recordings were identified using CellReg,⁴² an open-sourced MATLAB package that enables the tracking of cells across multiple recording sessions using a probabilistic approach. CellReg-identified cells that were present in both sessions were manually verified based on spatial footprint and calcium-transient.

Indirect Calorimetry

Raw data was exported from the PhenoMaster software. Respiratory exchange ratio (RER) was calculated as the ratio of CO₂ production over O₂ consumption. Energy expenditure from carbohydrate and fat was calculated from RER and total energy expenditure. Data was aligned to the start of the pre-stimulation baseline period. Pre-stimulation, stimulation, and post-stimulation phase means were each averaged over the entire 1 hr phase.

Supplementary Material

2

Table S4. Statistical tests, n, and p-values for each panel, related to STAR Methods (“General Analysis”).

Document S1. Figures S1–S5, Tables S1–S3.

HIGHLIGHTS.

• The brain is a key intermediate for physiological improvement following exercise

• Exercise strengthens inputs to and increases the activity of VMH SF1 neurons

• Activation of VMH SF1 neurons following exercise is required to improve endurance

• Exogenous activation of VMH SF1 neurons following exercise enhances endurance gains