Treadmill exercise prevents stress-induced anxiety-like behaviors via enhancing the excitatory input from the primary motor cortex to the thalamocortical circuit

Abstract

Physical exercise effectively prevents anxiety disorders caused by environmental stress. The neural circuitry mechanism, however, remains incomplete. Here, we identified a previously unrecognized pathway originating from the primary motor cortex (M1) to medial prefrontal cortex (mPFC) via the ventromedial thalamic (VM) nuclei in male mice. Besides anatomical evidence, both ex vivo and in vivo recordings showed enhanced excitability of M1-VM inputs to the prelimbic (PrL) region of mPFC upon 14-day treadmill exercise on a chronic restraint stress (CRS) mouse model. Further functional interrogations demonstrated that the activation of this neural circuit is both necessary and sufficient to direct the anxiolytic effect of exercise training in CRS mice. Our findings provide more insights into the neural circuits connecting motor and mental regions under exercise paradigm and implicate potential targets for neuromodulation in treating anxiety disorders.

Physical exercise relieves anxiety disorders, while the neural circuit remains incomplete. Here, authors show that treadmill training enhanced input from the primary motor region to the medial prefrontal cortex, via the ventromedial thalamic nuclei.

Introduction

Anxiety disorder is a prevalent mental illness worldwide and may induce both psychiatric and body symptoms¹. The high incidence of anxiety among different age groups^2–4, plus more than 10% prevalence of subthreshold anxiety disorder in the whole population⁵ has drawn the attentions of both psychiatric clinicians and neuroscientists. Although generations of drugs have been developed to relieve anxiety disorder, adverse effects^6,7 or drug addiction⁸ restrict their further applications. Therefore, non-drug approaches are required to improve the overall efficiency of treatment. Physical exercise is one effective measure to attenuate anxiety disorders in human studies^9–12. In rodent models of chronic restraint stress (CRS), our group recently identified the prevention of anxiety-like behaviors by persistent treadmill exercise training^13–15. The investigation of the neurobiological mechanism has revealed the involvement of neurogenesis^16,17, synaptogenesis^18,19, oxidative stress²⁰, and neuroinflammation²¹ in exercise-mediated anxiolysis, whilst leaving the potentially neural circuitry mechanism largely unattended.

Recent studies have revealed a complex neural circuitry network in the occurrence of anxiety disorders. Among major brain regions involved, the medial prefrontal cortex (mPFC) has been recognized to mediate the anxiety-like behaviors in rodents under chronic stress, via its specific projection to amygdala nuclei^22,23, hippocampus^24,25 or thalamic nuclei²⁶. Our previous findings have suggested the enhanced mPFC neuronal activity by treadmill exercise, in conjunction with improved stress resilience^13–15. It is thus interesting to investigate which neural circuit modulates mPFC activity during exercise training. Among the upstream imput regions of mPFC, amygdala^27,28, hypothalamus²⁹, ventral hippocampus³⁰ and ventral tegmental area (VTA)³¹ all present disrupted neural connectivity with mPFC during anxiety occurrence. However, current knowledge has only reported changes in dendritic spine morphology in hippocampus and mPFC after chronic exercise³², and has not attributed specific neural circuit that may transduce body movement to mental status. We thus hypothesize that some neural pathways may exist to connect motor regions with specific nuclei for mental functions, and the functional integrity of such pathways may help to explain the exercise effects in alleviating anxiety-like behaviors.

To test this hypothesis, we generated a mouse CRS model coupled with daily treadmill exercise. The 14-day physical training effectively prevented stress-induced anxiety-like behaviors, via potentiating the projection of ventromedial thalamus (VM) to mPFC. Further investigations revealed enhanced excitatory inputs from primary motor cortex (M1) to this VM-mPFC pathway upon exercise training. In sum, this M1-VM-mPFC provides a complete neural circuit for illustrating the mechanism of exercise in attenuating anxiety disorders.

Results

Physical exercise activates the ventromedial thalamic input to prelimbic region to prevent anxiety-like disorders

We first generated a mouse anxiety model via 14-day CRS treatment on adult (7-8 weeks) C57 mice. Concurrent application of 1-hour daily treadmill exercise effectively prevented the occurrence of anxiety-like behaviors as suggested by the open field and elevated plus-maze assays, while the general motor behavior was unaffected by either the CRS or exercise training (Fig. S1). On naïve, unstressed mice, however, exercise training itself did not affect the anxiety-like behaviors (Fig. S1), suggesting the selective enhancement of stress resilience. To characterize the possibly neural circuitry mechanism, major inputs to prelimbic (PrL) region of medial prefrontal cortex was determined to include both M1 and VM (Fig. 1a). However, when the M1-PrL pathway was blocked via specific ablation of PrL-projecting M1 neurons (Fig. S2a, b), physical exercise can still alleviate anxiety-like behaviors in CRS mice (Fig. S2c–f), suggesting the irrelevance of this M1-PrL pathway in anxiolytic effects. We thus studied the potential VM-PrL pathway. First, the excitability of PrL-projecting VM neurons was quantified using ex vivo electrophysiology (Fig. 1b, c), which showed an increased number of spikes and higher membrane excitability after exercise training (Fig. 1d–f). Moreover, exercise effectively enhanced the frequency of miniature excitatory postsynaptic current (mEPSC) of CRS mice, whilst leaving the amplitude unchanged (Fig. 1g-i). These results suggest enhanced excitatory inputs towards these PrL-projecting VM neurons under exercise training.

Fig. 1. Treadmill exercise activates prelimbic-projecting ventromedial thalamic neurons to prevent stress-induced anxiety-like behaviors.

a Left, viral injection sites; Right, Retrograde labelling of neurons in the primary motor cortex (M1) and the ventromedial thalamus (VM). PrL, prelimbic; CPU, caudate putamen. Green, EGFP. Scale bar, 100 μm (left- and right-most) and 500 μm (middle panel). b Timelines of experimental design. c Left, viral injection sites; Right, distribution of PrL-projecting neurons in VM, and patch-clamp recording of one representative cell. Green, EGFP. Scale bar, 100 μm (left) and 20 μm (right). d Sample spikes of VM neurons giving fixed (240 pA) injection currents. e Exercise training (Ex) elevates the total number of spikes in CRS mice. Two-way ANOVA with respect to the group effect, F(2,525) = 15.40, P < 0.0001. f Left, CRS+Ex animals showed elevated resting membrane potential (RMP). One-way ANOVA, F(2,33) = 5.472, P = 0.0089. Right, CRS+Ex animals displayed lowered rheobase value. F(2,33) = 5.562, P = 0.0083. n = 12 neurons from 3 mice in each group in (e, f). g Representative trances of miniature excitatory postsynaptic currents (mEPSCs) in all groups. h Distribution of mEPSC frequency. Exercise training enhanced the frequency of mEPSC in CRS-treated animals. F(2,36) = 13.49, P < 0.0001. i Distribution of mEPSC amplitude. Exercise did not affect mEPSC amplitude. One-way ANOVA, F(2,36) = 0.1044, P = 0.9011. n = 13 neurons from 3 mice in each group in (h, i). j Timelines of experimental design. k Left, viral injection sites. Right, distribution of PrL-projecting neurons in VM. Red, mCherry. Scale bar, 100 μm. l Chemogenetic manipulation did not change overall locomotor ability in the open field. F(2,21) = 0.05169, P = 0.9497. m The inhibition of PrL-projecting VM neurons decreased time spent in the central region. F(2,21) = 6.873, P = 0.0051. n CNO infusion did not affect the total distance travelled on the elevated plus-maze. F(2,21) = 1.799, P = 0.1900. o Chemogenetic inhibition of PrL-projecting VM cells decreased the time spent in the open arm region. F(2,21) = 7.292, P = 0.0039. N = 8 mice each group in (l–o). Tukey’s multiple comparison test was employed to make comparisons between two specified groups in a two-sided manner. All data were presented as mean ± sem.

To demonstrate the causal relationship between VM-PrL pathway and anxiolysis of treadmill exercise, we employed chemogenetic approach to specifically activate these PrL-projecting VM neurons via expressing excitatory receptor hM3Dq (Fig. S3a–d). The infusion of specific ligand clozapine-N-oxide (CNO) did not change total distance but markedly increased time duration in the central region or the open arms in CRS mice (Fig. S3e–h), just like those in exercised animals. Alternatively, when PrL-projecting VM neurons were inhibited in exercised group by specific expression of hM4Di (Fig. 1j, k), CNO infusion effectively abolished exercise effect as suggested by the reoccurrence of anxiety-like behaviors (Fig. 1l–o). This VM-PrL pathway, however, only relieved the anxiety-like behaviors but was not anxiogenic itself, as suggested by the chemogenetic inhibition approach in unstressed individuals (Fig. S3i–n). Therefore, exercise training potentiates VM inputs to cortical regions for conferring stress resilience.

Physical exercise potentiates neuronal activity in PrL region to confer stress resilience

Given the enhanced inputs from VM to PrL, we next investigated the postsynaptic change of PrL neurons receiving VM projection. Using in vivo neuronal activity of PrL using genetically coded fluorescent calcium sensor GCaMP7f and 2-photon recording (Fig. 2a–c). The quantification of calcium transients in PrL neurons receiving VM inputs showed down-regulated neuronal activities under CRS (Fig. 2d–f), as consistent with our recent reports in PrL^13,15. Moreover, exercise training elevated total calcium strength in CRS mice, including higher peak value and frequency of calcium spikes (Fig. 2d–f).

Fig. 2. The activation of VM-PrL pathway is necessary for exercise-mediated anxiolytic effects.

a Schematic illustration of virus injection. b Left, fluorescent images of viral infection. Green, GCaMP. Scale bar, 300 μm. Right, the field of view (FOV) of in vivo 2-photon imaging. Scale bar, 100 μm. c Heatmaps of individual in vivo neuronal calcium activities during a 150-sec recording window. A total of 40 sampled neurons from 4 mice were plotted in each group. d Exercise training potentiated the total integrated calcium strength. Nonparametric Kruskal-Wallis test statistic=39.27, P < 0.0001. e Exercise training increased the peak value of individual calcium transient. Nonparametric Kruskal-Wallis test statistic=41.74, P < 0.0001. f Distribution of calcium transient frequency. n = 133 neurons from 4 mice in each group in (d–f). g Timelines of experimental design. h Left, viral injection sites. Right, distribution of VM innervating neurons in PrL. Scale bar, 250 μm. Red, mCherry. i Chemogenetic manipulation did not change overall locomotor ability in the open field. One-way ANOVA, F(2,21) = 0.5787, P = 0.5693. j The inhibition of VM innervating PrL neurons decreased time spent in the central region. F(2,21) = 5.903, P = 0.0092. k CNO infusion did not affect the total distance travelled on the elevated plus-maze. F(2,21) = 0.1739, P = 0.8416. l Chemogenetic inhibition of VM innervating PrL cells decreased the time spent in the open arm region. F(2,21) = 7.363, P = 0.0038. N = 8 mice each group in (i–l). The comparison between two specified groups was performed using Dunn’s multiple comparison test in a two-sided manner in (d–e), and Tukey’s multiple comparison test in a two-sided manner in (i–l). All data were presented as mean ± sem.

Based on such in vivo recording results, chemogenetic approaches were further employed to testify the involvement of VM-PrL pathway in exercise-mediated anxiolysis. The selective activation of PrL neurons at the downstream of VM (Fig. S4a, b) mimicked the effects of exercise to improve the stress resilience of animals (Fig. S4c–f). On the other hand, when these VM-innervating PrL neurons was specifically inhibited in exercised mice (Fig. 2g, h), the anxiolytic effect of animals by exercise training was largely abolished (Fig. 2i–l). Taken together, exercise training prevents anxiety-like behaviors under CRS possibly via activating the VM-PrL pathway.

The M1 projects to VM for attenuating stress-induced anxiety-like behaviors

After establishing the role of VM-PrL pathway in stress resilience by exercise training, we further tracked the possible inputs of VM nuclei. We mainly focused on brain regions involving in motor control. Using a retrogradely labelling approach, both M1 and deep cerebellar nuclei (DCN) showed inputs to VM nuclei (Figs. 3a, S5). The DCN-VM pathway, however, did not participate in the anxiolytic effect of exercise training as shown by our chemogenetic inhibition assay (Fig. S5) and by a recent work, which showed the mediation of the central amygdala nuclei by cerebellar nuclei for alleviating anxiety-like behaviors³³. Using a circuit-specific cell ablation approach to remove VM-projecting M1 neurons (Fig. 3b, c), we found that although the overall motor activity remained unchanged (Fig. 3d, f), exercise training could not prevent anxiety-like behaviors (Fig. 3e, g). These results illustrated the necessity of M1 projection to VM for exercise-conferred stress resilience.

Fig. 3. M1 neurons project to VM for improving anxiety disorders.

a Left, schematic illustration of viral injection; Right, Retrograde labelling of neurons in M1 across the anteroposterior axis. M1, primary motor cortex; M2, secondary motor cortex; DCN, deep cerebellar nuclei. Green, GFP. Scale bar, 500 μm. b Upper, timelines of experimental design. Lower left, viral injection sites; Lower right, distribution of VM-projecting neurons in PrL. S1, primary sensory cortex; Scale bar, 250 μm. c The quantification of VM-projecting cell number in M1 after cell ablation. Multiple t-test was used for the comparison in a two-sided manner. N = 6 mice per group. d The ablation of VM-projecting M1 neurons did not change overall locomotor ability in the open field. One-way ANOVA, F(2,21) = 0.4272, P = 0.6578. e The specific cell ablation decreased time spent in the central region. F(2,21) = 9.225, P = 0.0013. f Cell ablation in M1 did not affect the total distance travelled on the elevated plus-maze. F(2,21) = 0.2087, P = 0.8133. g The blockade of M1-VM pathway decreased the time spent in the open arm region. F(2,21) = 8.785, P = 0.0017. N = 8 mice each group in (d–g). h Timelines of experimental design. i Left, viral injection sites. Right, distribution of VM-projecting neurons in M1. Red, mCherry. Scale bar, 100 μm. j Current-evoked action potentials in a representative VM-projecting M1 neuron labelled with hM4Di recorded before, during and after CNO perfusion (10 μM). k CNO infusion effectively repressed cellular excitability. One-way ANOVA with repeated measures, F(1.241,6.207) = 93.14, P < 0.0001. n = 6 neurons from 3 mice. l Chemogenetic manipulation did not change overall locomotor ability in the open field. F(2,24) = 0.8324, P = 0.4472. m The inhibition of VM-projecting M1 neurons decreased time spent in the central region. F(2,24) = 26.81, P < 0.0001. n CNO infusion did not affect the total distance travelled on the elevated plus-maze. F(2,24) = 0.9322, P = 0.4075. o Chemogenetic inhibition of PrL-projecting VM cells decreased the time spent in the open arm region. F(2,24) = 6.958, P = 0.0041. N = 9 mice each group in (l–o). Tukey’s multiple comparison test was employed to make comparisons between two specified groups in a two-sided manner. All data were presented as mean ± sem.

Given the potentiation of cortical neuronal activity under exercise training¹⁸, we next manipulated the cortical neural activity coupled with behavioral assays to establish functional evidence. The inhibition of the VM-projecting M1 neurons using inhibitory chemogenetic receptors (Fig. 3h–k) in exercised animals resulted in the reoccurrence of anxiety-like behaviors (Fig. 3l–o). In a similar manner, the excitation of these M1 neurons in CRS-treated animals (Fig. S6a–d) effectively relieved anxiety-like behaviors (Fig. S6e–h). These results further supported the necessary role of M1 inputs to VM in exercise-mediated anxiolysis.

We next investigated the activity of VM neurons receiving M1 inputs. By combining anterograde transsynaptic labelling from M1 and electrophysiological recording (Fig. 4a, b), we observed elevated membrane excitability of VM neurons under exercise training, whilst the CRS treatment had no effect (Fig. 4c–e). The recording of synaptic transmission also implied potentiated inputs from M1 to VM neurons by treadmill exercise (Fig. 4f–h). Such ex vivo observations paralleled our initial findings, in which PrL-projecting VM cells presented higher excitability upon exercise training (Fig. 1b–i), suggesting a possible connection between M1 and PrL via VM nuclei.

Fig. 4. VM nuclei is activated by M1 for relieving anxiety-like behaviors.

a Timelines of experimental design. b Left, schematic illustration of viral injection; Right, distribution of M1 innervating neurons in VM, and patch-clamp recording of one representative cell. Red, mCherry. Scale bar, 100 μm (left) and 20 μm (right). c Sample spikes of VM neurons giving fixed (240 pA) injection currents. d Exercise training (Ex) elevates the total number of spikes in CRS mice. Two-way ANOVA with respect to the group effect, F(2,495) = 34.52, P < 0.0001. e Left, CRS+Ex animals showed elevated resting membrane potential (RMP). One-way ANOVA, F(2,36) = 3.972, P = 0.0276. Right, CRS+Ex animals displayed lowered rheobase value. F(2,36) = 9.229, P = 0.0006. n = 13 neurons from 3 mice in each group in (d, e). f Representative trances of miniature excitatory postsynaptic currents (mEPSCs) in all groups. g Distribution of mEPSC frequency. Exercise training enhanced the frequency of mEPSC in CRS-treated animals. F(2,33) = 4.362, P = 0.0209. h Distribution of mEPSC amplitude. Exercise did not affect mEPSC amplitude. F(2,33) = 0.9798, P = 0.3857. n = 12 neurons from 3 mice in each group in (g, h). Tukey’s multiple comparison test was employed to make comparisons between two specified groups in a two-sided manner. All data were presented as mean ± sem.

The M1-VM-PrL pathway mediates exercise effects in relieving anxiety-like behaviors

To identify the existence of the M1-VM-PrL pathway, we utilized both retrograde and anterograde labelling strategy and observed prominent co-localization in VM nuclei, which largely belonged to excitatory CaMKIIα neurons (Fig. 5a–c). The identity of this pathway was later studied using ex vivo optogenetic stimulation of M1-innervated VM axonal terminus in PrL (Fig. 5d). Patch clamp recording of adjacent PrL neurons identified rapid depolarization (Fig. 5e). The abolishment of such action potential by tetrodotoxin (TTX), plus the recovery under 4-aminopyridine (4-AP) infusion suggested the monosynaptic excitatory connection between M1-VM pathway and PrL cells (Fig. 5e, f). A follow-up study recorded in vivo calcium activity of axonal terminus of M1-innervated VM neurons in PrL on awake, head-fixed mouse (Fig. 5g, h) and found potentiated axonal activities under exercise training (Fig. 5i-l). Such M1 neurons project to the whole VM (Fig. S7a, b), and these VM neurons innervated by M1 showed similar spatial distribution patterns as these of PrL-projecting VM cells (Fig. S7c, d). Functional evidence showed that M1 inputs were necessary to maintain the normal excitability of PrL-projecting VM neurons in exercised mice (Fig. S8), providing the explanation for higher neuronal calcium activities of PrL cells at the downstream of VM (Fig. 2a–f) and highlighting the potentiation of M1-VM pathway to provide excitatory inputs to PrL cells under exercise training.

Fig. 5. Exercise training potentiates excitatory input to PrL by M1-VM pathway.

a Viral injection schemes. b Fluorescent colabelling of antero- and retro-grade labelling VM neurons. FrA, frontal association cortex; PrL, prelimbic region; MO, medial orbital cortex; LO, lateral orbital cortex; VO, ventral orbital cortex; ZI, zona incerta; VM, ventromedial thalamus; Sub, subcoeruleus nucleus; mt, medial terminal nucleus of the accessory optic tract. Scale bar, 150 μm and 25 μm (enlarged views only). c Percentage of excitatory (CaMKIIα) neurons in co-labelled VM neurons. N = 6 mice. d Left, viral injection and ex vivo optogenetic stimulation schemes. Right, fluorescent images showing viral infection in VM. Scale bar, 250 μm. e Representative traces showing EPSCs when slices were sequentially perfused with ACSF (left), TTX (1 µM, middle left), TTX + 4-AP (100 µM, middle right) and TTX + 4-AP + CNQX (20 µM, right). f Summary plots of the EPSC amplitudes in (e). One-way ANOVA with repeated measures, F(1.167,7.002) = 93.14, P = 0.0004. n = 7 neurons from 3 mice. g Viral injection schemes. h Fluorescent images showing viral infection in VM and PrL. DLO, dorsolateral orbital cortex Green, GCaMP. Scale bar, 500 μm (left) and 100 μm (right). i Heatmaps of in vivo calcium activities of axonal terminals during a 150 s recording window. A total of 20 sampled axons from 4 mice were plotted in each group. j Exercise training potentiated the total integrated calcium strength. Nonparametric Kruskal-Wallis test statistic=111.2, P = 0.0001. k Exercise training also increased the peak value of individual calcium transient. Nonparametric Kruskal-Wallis test statistic=101.0, P < 0.0001. l Distribution of calcium transient frequency. n = 116 neurons from 4 mice in each group in (j–l). The comparison between two specified groups was performed using Tukey’s multiple comparison test in a two-sided manner in (f), and Dunn’s multiple comparison test in a two-sided manner in (j, k). All data were presented as mean ± sem.

Next, the functional role of this pathway in anxiolysis was investigated. By optogenetic inhibition of axonal terminus of M1-innervated VM neurons within the PrL at 4 h before behavioral tests (Fig. 6a–c), exercised mice presented anxiety-like behaviors (Fig. 6d–g), suggested the elimination of anxiolytic effect of exercise training. In a second cohort of assays, M1-innervated, PrL-projecting VM neurons were labelled with hM4Di using a combination of Cre and Flp system (Fig. 6h, i). The chemogenetic inhibition of these relaying VM neurons also deprived exercise effects in preventing anxiety-like behaviors (Fig. 6j–n). These results highlighted the necessary role of the M1-VM-PrL circuit in exercise-mediated anxiolysis.

Fig. 6. The M1-VM-PrL pathway mediates exercise effect in relieving anxiety-like behaviors.

a Timelines of experimental design. b Left, viral injection sites. Right, fluorescent images showing viral infection in VM and PrL. Green, EGFP. Scale bar, 150 μm. c Sample spikes of VM neurons giving fixed (300 pA) injection currents plus optogenetic inhibition. d Light inhibition (eNpHR3) did not change overall locomotor ability in the open field. One-way ANOVA, F(2,27) = 0.6808, P = 0.5147. e Specific light inhibition of M1-VM terminus in PrL region decreased time spent in the central region. F(2,27) = 7.535, P = 0.0025. f The optogenetic manipulation did not affect the total distance travelled on the elevated plus-maze. F(2,27) = 0.9190, P = 0.4110. g The inhibition of M1-VM terminus decreased the time spent in the open arm region. F(2,27) = 5.352, P = 0.0110. N = 10 mice each group in (d–g). h Timelines of experimental design. i Left, viral injection sites. Right, fluorescent images showing viral infection in VM. Green, EGFP Scale bar, 250 μm. j Current-evoked action potentials in a representative M1-innervated, and PrL-projecting VM neurons were labelled with hM4Di recorded before, during and after CNO perfusion (10 μM). k Chemogenetic inhibition (CNO group) did not change overall locomotor ability in the open field. One-way ANOVA, F(2,21) = 0.1865, P = 0.8312. l CNO infusion decreased time spent in the central region. F(2,21) = 6.876, P = 0.0050. m The chemogenetic manipulation did not affect the total distance travelled on the elevated plus-maze. F(2,21) = 0.5705, P = 0.5737. n The chemogenetic inhibition of M1-VM-PrL pathway decreased the time spent in the open arm region. F(2,21) = 4.975, P = 0.0170. N = 8 mice each group in (k–n). Tukey’s multiple comparison test was employed to make comparisons between two specified groups in a two-sided manner. All data were presented as mean ± sem.

Last, whether this pathway is sufficient to maintain stress resilience was tested by optogenetics excitation of M1-VM axonal inputs to PrL region (Fig. 7a–c). Behavioral assays showed that a single light stimulation session was sufficient to suppress anxiety-like behaviors in CRS-treated, exercise-free mice (Fig. 7d–g). Since these PrL-projecting VM neurons mainly innervate PrL, with less distribution of collateral projection to other brain regions (Fig. S9), it is believed that the anxiolytic effect was mainly mediated via this M1-VM-PrL pathway. Moreover, the chemogenetic excitation of the specific VM neuronal subpopulation that both receives inputs from M1 and projects to PrL (Fig. 7h–j) resulted in improved stress resilience (Fig. 7k–n). These results agreed with earlier assays on PrL-projecting VM nuclei in anxiolysis (Fig. S3). Taken together, our anatomical, physiological and behavioral evidence converged to suggest that treadmill exercise potentiates M1-VM inputs to PrL neurons, improving the stress resilience.

Fig. 7. Activation of M1-VM-PrL pathway relieves stress-induced anxiety-like behaviors.

a Timelines of experimental design. b Left, viral injection sites. Right, M1-inneravting VM neurons and their terminus in PrL. Green, EGFP Scale bar, 100 μm. c Sample spikes of PrL neurons under light stimulation. d Light activation (ChR2) did not change overall locomotor ability in the open field. One-way ANOVA, F(3,32) = 0.7786, P = 0.9956. e Specific light activation of M1-VM terminus in PrL region increased time spent in the central region. F(3,32) = 6.508, P = 0.0015. f The optogenetic manipulation did not affect the total distance travelled on the elevated plus-maze. F(3,32) = 0.07647, P = 0.9674. g The activation of M1-VM terminus increased the time spent in the open arm region. F(3,32) = 5.067, P = 0.0055. N = 9 mice each group in (d–g). h Timelines of experimental design. i Left, viral injection sites. Right, fluorescent images showing viral infection in VM. Green, EGFP Scale bar, 250 μm. j Sample traces of VM neurons giving under CNO infusion. k Chemogenetic activation (CNO group) did not change overall locomotor ability in the open field. F(2,21) = 0.09855, P = 0.9066. l CNO infusion increased time spent in the central region. F(2,21) = 7.931, P = 0.0027. m The chemogenetic manipulation did not affect the total distance travelled on the elevated plus-maze. F(2,21) = 0.2438, P = 0.7858. n The chemogenetic activation of M1-VM-PrL pathway increased the time spent in the open arm region. F(2,21) = 8.759, P = 0.0017. N = 8 mice each group in (k–n). Tukey’s multiple comparison test was employed to make comparisons between two specified groups in a two-sided manner. All data were presented as mean ± sem.

Discussion

In the current work, we reported a previously unrecognized neural pathway from M1 to PrL, in which VM acts as the relaying nuclei. This excitatory circuit was selectively activated upon treadmill exercise and resulting in enhanced neuronal activities of PrL neurons for preventing anxiety-like disorders. Although the VM-PrL circuit itself did not show any significant roles in the pathogenesis of anxiety-like behaviors (Fig. S3i–n), its activation by exercise-mediated M1 input did present an anxiolytic role. These findings were consistent with our previous findings showing the activation of M1 neurons under chronic treadmill training^18,19, and provide anatomical and physiological evidence to connect motor regions with mental-related nuclei, thus illustrating the brain network changes during exercise regime.

For a long time, the mechanistic studies of exercise in relieving anxiety or depressive disorders mainly focused on the improvement of neurogenesis^17,34 or synaptogenesis^35,36, while leaving the neural circuitry changes largely unknown. Our knowledge of long-term effects of exercise training on neural circuitry adaptation mainly come from the motor circuit itself. For example, aerobic exercise can facilitate the neural plasticity of motor network in Parkinson’s disease (PD) patients³⁷, and treadmill training helped to improve the synaptogenesis of motor cortex^18,19. For neural circuits involved in mental functions, recently human brain studies have reported enhanced baseline salience circuitry connectivity upon exercise, in conjunction with improved anxiety symptoms³⁸. However, few studies have been provided in animal models and we thus investigated the detailed neural mechanism using a rodent model.

The anatomical connection between M1 and VM has already been reported. In previous studies, pyramidal neurons in anterolateral motor cortex form bidirectional connection with VM nuclei to assist the information processing within the motor system³⁹. Moreover, DCN neurons project to VM for modulating associative sensorimotor response⁴⁰. The VM nuclei was thus recognized for motor behavioral regulations⁴¹ and acted as one of the salient hub regions within the brain during the task-related brain network reorganization⁴². In addition to receiving motor inputs, VM is also innervated by various mental-related brain regions including mPFC^43,44 and hypothalamus⁴⁵. In another study, cerebellar nuclei projects to mPFC via VM for mediating social deficits and repetitive behaviors in autism model mice⁴⁶. This evidence extends the function of VM beyond simply motor regulation and provides more possibilities for its involvement in the regulation of different aspects of mental functions.

The mPFC forms bidirectional connection with VM, as supported by the driving of thalamic neurons from mPFC, in addition to the wide distribution of thalamic-cortical neurons in VM⁴⁴. A recent study reported the engagement of VM neurons in modulating neuron-derived neurotrophic factor (NDNF + ) cells in L1a of mPFC⁴⁷. Our results, however, identified VM-innervated PrL neurons to be distributed across all cortical layers (Fig. 2h). This finding agreed with previous observations showing the distribution of VM terminals in cortical layer 1, 3 and layer 5/6⁴⁸ and suggested the modulation of both interneurons and pyramidal neurons in PrL by VM projection⁴⁹. Since PrL pyramidal neurons are involved in anxiety-like behaviors⁵⁰ and the activation of PrL excitatory neurons relieved anxiety⁵¹, the enhanced excitatory inputs from M1-VM to PrL neurons provides the neural substrate for improving stress resilience under exercise scheme. It is further noticed that even for the PrL, the heterogeneity of neurons make it is unlikely to simply correlate the anxiety status with the neural activity of the whole region. In our latest report, for example, exercise training inhibits the BLA-projecting PrL neurons for relieving the anxiety-like behaviors⁵². In CRS animals, our previous works also showed the depressed calcium activities in PrL^13–15, which also observed in our model (Fig. 2a–f). We thus propose that such discrepancy reflect the complexity of PrL neural network.

The possible involvement of M1-VM-PrL pathway in exercise-mediated anxiolytic effect may have paralleled results in human brain imaging studies. For example, the transcranial direct current stimulation (tDCS) on M1 resulted in the increased response of the thalamus⁵³, and transcutaneous vagus nerve stimulation (tVNS) induces neural activity in mPFC and thalamus under functional MRI⁵⁴. It is interesting to further investigate the downstream targets of those PrL neurons receiving M1-VM inputs in our model. Such PrL pyramidal cells are unlikely to project to amygdala nuclei, as the elevated prelimbic-to-BLA activity is usually associated with higher anxiety levels^50,55. Therefore, other PrL-projecting regions such as the nucleus accumbens (NAc) and VTA may help to explain the anxiolytic effects of exercise, as such corticoaccumbal and corticotegmental pathways were all involved in anxiety-like behaviors³¹. Future studies can be pursued to identify the downstream target of this cortico-thalamic-cortical pathway under exercise regime.

Exercise, like other types of life-style interventions, may affect the neuroplasticity to change neural circuits involved in cognition and mental functions⁵⁶. Our previous work supports this model by showing the elevation of blood-borne factors including those affect RNA methylation of synaptic transcripts¹³, the lactylation of synaptic proteins¹⁵, and cytokine that inhibit neuroinflammation⁵⁷. However, these factors tend to act in a brain-wide manner, with the lack of region-specificity. The current work identified the role of M1 in exercise-induced stress resilience. Since we previously showed the continuous activation of M1 projecting neurons under treadmill exercise¹⁸, the upregulation of excitatory inputs to the downstream VM-PrL pathway can be attributed to exercise training. In human studies, clinicians have reported various effects of repetitive transcranial magnetic stimulation (rTMS) on M1 beyond conventional motor rehabilitation, including the improvement of anxiety or depressive disorders^58,59. Our works on animal models thus provide mechanistic explanations for the intervention of mental disorders via targeting the motor region.

In sum, treadmill exercise potentiates the excitatory input to PrL via modulating the M1-VM neural circuit, to prevent anxiety-like disorders upon chronic stress. This model establishes a previously unrecognized neural circuit that connects motor regions with mental nuclei and provides targets for neuromodulation in alleviating anxiety disorders, in addition to the mechanistic interpretation of exercise intervention on psychiatric diseases.

Methods

Experimental animals

Male C57BL/6 J mice (5 weeks old) were purchased from the Guangdong Medical Experimental Animal Center. All animals were housed in standard cages, five in a cage, with ad libitum access to food and water, 40–60% humidity, 21–25 °C, and a 12 h light/dark cycle (light: 8:00 a.m. to 8:00 p.m.).

In the chronic restraint stress (CRS) model, mice were placed in a plastic pastry bag and restrained for 3 h per day (8:00 p.m. to 11:00 p.m.) for fourteen days. For the exercise model, mice were placed on a treadmill (Zhongshi Technology, China) at a speed of 10 m/min for 1 h (10:00 a.m. to 11:00 a.m.) continuously for 14 days, while the control group and mice in the CRS group were placed on the treadmill track at the same time every morning and allowed to move freely. All animal experimental protocols were pre-approved by the Ethics Committee of Experimental Animals of Jinan University in accordance with Institutional Animal Care and Use Committee guidelines for animal research.

Stereotactic injection

Mice were anesthetized with tribromoethanol, fixed on a rack, and placed in a stereotaxic apparatus. Erythromycin eye ointment was used to smear the mouse eyes to prevent corneal dryness. A high-speed skull drill was then used to drill a hole, and a glass microelectrode was installed on the micro injection pump. The controller was adjusted to inject the virus at a speed of 50 nL/min. At the end of the virus injection, the needle was left in place for five minutes to allow the virus to fully diffuse.

To separately label neurons projecting from M1 to VM and from VM to PrL, we injected AAV-Retro-EGFP (virus titers: 5.4 × 10¹² GC/mL, 120 nL) virus into VM (AP: −0.95 mm, ML: ±0.9 mm, DV: −4.2 mm) and PrL (AP: +2.6 mm, ML: ±0.15 mm, DV: −1.6 mm).

Using chemical genetics to control neurons projecting from M1 to VM and neurons projecting from VM to PrL, AAV2/2-Retro-Cre (virus titers: 3 × 10¹² GC/mL, 100 nL) was injected into VM and PrL of C57BL/6 mice, and AAV2/9-DIO-mCherry, AAV2/9-DIO-hM3Dq-mCherry, or AAV2/9-DIO-hM4Di-mCherry (virus titers: 3.5 × 10¹² GC/ml, 150 nL) was injected into M1 (AP: +1.4 mm, ML: ±1.5 mm, DV: −1.6 mm) and VM.

To specifically infect VM neurons receiving M1 projections and PrL neurons receiving VM projections, we injected AAV2/1-Cre (virus titers: 1.5 × 10¹³ GC/ml, 100 nL) or AAV2/9-hEF1a-WGA-Cre-WPRE-PA (virus titers: 1.5 × 10¹³ GC/ml, 100 nL) into M1 and VM, respectively. AAV2/9-DIO-EGFP (virus titers: 3.5 × 10¹² GC/ml, 150 nL) was then injected into VM and PrL, respectively.

To specifically infect relay neurons in the VM, AAV2/1-Cre (virus titers: 1.5 × 10¹³ GC/ml, 100 nL) or AAV2/9-hEF1a-WGA-Cre-WPRE-PA (virus titers: 1.5 × 10¹³ GC/ml, 100 nL) was injected into M1, AAV2/2-Retro-FLP (virus titers: 2 × 10¹² GC/ml, 100 nL) into PrL, AAV-Con/Fon-hM3Dq-EGFP or AAV-Con/Fon-eNpHR3-EGFP (virus titers: 2 × 10¹² GC/ml, 150 nL) into the VM.

All viral information is listed in Table S1.

Ex vivo electrophysiological recording

Mice were deeply anesthetized with isoflurane and decapitated, and their brains were quickly removed and coronal slices (250 μm) containing the PrL were cut out by a VT1000S Vibratome (Leica Microsystems, Wetzlar, Germany) using ice-cold, sucrose-based artificial cerebrospinal fluid (sACSF) saturated with 95% O₂/5% CO₂ (carbogen). The sACSF contained (in mM): 64 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 10 MgCl₂, 0.5 CaCl₂, 26 NaHCO₃, 10 glucose, and 120 sucrose. Slices were then transferred to pre-warmed (32°C) ACSF containing (in mM): 126 NaCl, 2.5 KCl 1.2 NaH₂PO₄, 10 glucose, 26 NaHCO₃, 2.4 CaCl₂, and 1.2 MgCl₂, pH 7.4, 295 mOsm, and oxygenated with 95% O₂/5% CO₂. After 30 min at 32°C and an additional 30 min at room temperature, recordings were made. Recording electrodes, with a resistance of 4–6 MΩ, were pulled from filamented borosilicate glass capillaries (inner diameter, 0.86 μm) using a P-97 horizontal pipette puller (Sutter Instrument Co., Novato, CA). They were filled with intracellular solution containing 135 mM K-gluconate, 5 mM KCl, 10 mM HEPES, 0.2 mM EGTA, 4 mM MgATP, 10 mM Na₂-phosphocreatine, and 0.3 mM Na₃GTP, pH adjusted to 7.4 with KOH.

To assess the neuronal excitability, the glass electrode was loaded with an intracellular solution. Sequential depolarizing currents, incremented by 30 pA steps and each lasting for 1000 ms, were injected into the neurons. When recording from neurons in the M1 and VM areas, the current intensity ranged from −90 pA to 450 pA. For recordings in the PrL, the current intensity spanned from −90 pA to 300 pA. All recordings were performed in the current clamp mode.

To measure the miniature excitatory postsynaptic currents (mEPSCs), 1 μM tetrodotoxin (TTX) and 20 μM bicuculline were added to the artificial cerebrospinal fluid, and the cells were clamped at -70 mV under voltage clamp.

Blue light stimulation of axon terminals of neurons infected with AAV-ChR2 virus can induce postsynaptic currents. Brain slices were perfused with TTX (1 μM), 4-aminopyridine (4-AP, 100 μM), and CNQX (20 μM) to investigate the presence of direct synaptic connections between neurons. Under normal ACSF perfusion conditions, we randomly recorded neurons in the PrL that exhibited light-responsive activity. Light intensity was adjusted to attain EPSC amplitudes within the 50-200 pA range. PPR calculation is based on the ratio of the amplitude of the second EPSC to that of the first EPSC.

In order to verify the functions of ChR2, eNpHR3, hM4Di, and hM3Dq, the responses of neurons expressing ChR2 were detected by blue light pulses of different frequencies (1–20 Hz, pulse width 5 ms) under whole-cell patch clamp current clamp, and the responses of neurons expressing ChR2 were detected by yellow light stimulation (1 s) to inhibit the evoked action potentials of eNpHR3-expressing neurons, and to record the evoked action potentials of neurons expressing hM4Di or hM3Dq before and after perfusion with CNO.

All recordings were conducted employing a Multiclamp 700B amplifier (Molecular Devices). Traces were filtered at 2 kHz and digitized at 10 kHz. Light stimulation was administered through digital commands from the Digidata 1550 A and Digital stimulator (PG4000a, Cygnus Technology). Once stable whole-cell recordings were established with an access resistance below 25 MΩ, baseline electrophysiological properties were documented. Offline data analysis was carried out using Clampfit 10.0 software (Molecular Devices).

Open field test

The open field experiment was carried out before the elevated plus-maze assay. The open field box was made of transparent, opaque plastic (50 cm × 50 cm) with an overhead video tracking system (Norders, The Netherlands). At the beginning of the experiment, it was divided into 16 small areas on the software, and the middle four areas were defined as the central area. Each mouse was gently placed in the central area and allowed to freely explore the environment for five minutes. The time the mice stayed in the central area and the total movement distance were analyzed using EthoVision v7.0.

For chemogenetic assays, CNO (1 mg/kg, i.p.) was injected at 2 h after each daily running or restraint stress. All behavioral assays were performed at 24 h after the conclusion of the last exercise training.

Elevated plus-maze

The same cohort of mice was used for the open field and elevated plus-maze assay. The elevated plus-maze consists of two closed arms (30 cm × 5 cm) and two open arms (30 cm × 5 cm) connected by a platform (5 cm × 5 cm) in the central area, approximately 75 cm above the ground. At the start of the experiment, each mouse was placed on the central area of the elevated maze, facing an open arm, and allowed to explore freely for 5 min. The mouse was only considered to have entered the open arm when all four feet were inside. The time spent in the open and closed arms, as well as the distance traveled, were then analyzed.

Immunofluorescent staining

Mice were anesthetized with tribromoethanol, and their hearts were injected with saline and 4% paraformaldehyde (PFA) using a syringe. The brain tissue was then fixed in PFA and dehydrated with 30% sucrose solution for 48 h. Coronal sections containing the region of interest (40 μm) were sectioned from the brain tissue using a slide microtome. These sections were then washed with PBS and blocked with CDC-blocking, followed by the addition of the primary antibody (Rabbit Anti-CaMKIIa, Abcam, Cat#ab5683; RRID: AB_305050; 1:500), which was incubated at 4 °C for 48 h. The primary antibody was then discarded, and the sections were washed again before the addition of the secondary antibody (DyLight 647 Goat Anti-Rabbit, Thermo Fisher Scientific, Cat#a-21244; RRID: AB_2535812; 1:300), which was incubated for 2 h. After washing with PBS, the slices were spread, and fluorescent images were taken with a confocal microscope. Antibody information was listed in Table S1.

Two-photon transcranial in vivo calcium imaging

To record calcium signals from neurons receiving VM projections in PrL, AAV2/9-hEF1a-WGA-Cre-WPRE-PA virus was injected into VM and AAV2/9-EF1a-DIO-GCaMP7f (virus titers: 2 × 10¹² GC/ml, 150 nL) virus into PrL. In order to record calcium signals at PrL axon terminals of neurons receiving M1 projections in VM, AAV2/1-Cre virus was injected into M1, and AAV2/9-EF1a-DIO-GCaMP6m virus was injected into VM. After the viruses had been expressed for two weeks, the modeling was started for two weeks. After the modeling was completed, two-photon imaging was performed on awake mice with their heads fixed.

First, the mice were anesthetized with tribromoethanol, the scalp was cut with scissors to expose the skull, and then 75% alcohol was used to disinfect and remove the connective tissue on the surface of the skull. A custom-made ring metal plate with extensions on both sides was then glued to the target skull with the target area in the middle of the ring. The perimeter of the metal ring was further reinforced with dental cement, and the mice were returned to their cages to recover overnight.

Before the formal experiment, the mouse’s head was fixed on an aluminum plate, and it received three 20 min adaptation training sessions. A window was then drilled for imaging with a skull drill, and the exposed dura was lifted. A medical sponge was used to absorb the oozing blood, and then the area was rinsed with artificial cerebrospinal fluid to ensure that the imaging window was clear. A custom-made circular coverslip (diameter: 2.3 mm) was then placed on the surface of the brain tissue and sealed with tissue glue. The two-photon imaging laser used a 920 nm laser, and the lens was soaked in artificial cerebrospinal fluid. Calcium activity in neurons or axon terminals in the PrL brain region was recorded at a frequency of two Hz for 150 s. Five target areas (size: 400 μm × 400 μm) were recorded for each animal.

Image videos from calcium imaging were corrected using ImageJ’s TurboReg. The region of interest (ROI) was manually selected using the ROI Manager, and the fluorescence intensity was calculated by averaging the pixels. Neurons that were overburdened or had no calcium signal were excluded first. All data were normalized (F0) processing; F0 is the lowest average fluorescence intensity within 150 s of recording, and its relative value calculation formula is ΔF/F0 = (F-F0)/F0 × 100%. The total calcium signal of a neuron was defined as the total output intensity within 150 s. Frequency was calculated as the number of spikes per minute, with those peaks exceeding the mean by at least two standard deviations being counted as a valid spike.

In vivo optogenetic manipulations

To stimulate relay neurons in the VM, AAV2/1-Cre virus was injected in M1, AAV2/2-Retro-FLP virus in PrL, and AV-Con/Fon-eNpHR3-EGFP in VM. To stimulate the PrL axon terminals of neurons receiving M1 projections in the VM, AAV2/1-Cre virus was injected in M1 and AAV2/9-DIO-ChR2-EGFP virus in VM. After two weeks of virus expression, a buried fiber was inserted in VM (AP: −0.95 mm, ML: ±0.9 mm, DV: −4 mm) and PrL (AP: +2.6 mm, ML: ±0.15 mm, DV: −1.3 mm). Briefly, mice were anesthetized, the scalp was cut to expose the skull, a small hole was drilled with a skull drill, and a custom optical fiber (diameter: 0.2 mm) was inserted. Then two screws were screwed into the skull and fixed with dental cement. After one week of recovery, two weeks of modeling began. Mice were then transferred to their home cages and allowed for a 30 min acclimatization period, followed by illumination with 1 Hz blue light pulses (2 ms in duration) or yellow light pulses (2 ms in duration) for 10 min delivered via the optical cannula, which was connected to the optical patch cable. Behavioral testing was performed after stimulation 4 h later.

Statistics and reproducibility

For the fluorescent images as representative panels in the figures, at least 3 consecutive slices were performed from one mouse, and no less than 3 mice from the same treatment group were checked for the validity.

Statistical analysis was performed using GraphPad Prism 9.5.1 (La Jolla, CA, USA). All data were presented as mean ± sem. Each dataset was tested for normality, and if the dataset passed the normality test, parametric tests were used. A two-sample Student’ t-test was used for comparisons between two groups, and one-way analysis of variance (ANOVA) with Tukey’s post hoc comparison was used for differences between multiple groups. For two independent variables, two-way ANOVA with Bonferroni post hoc comparisons were used. The Kruskal-Wallis test was used for multi-group comparison in non-parametric datasets. All data were analyzed by researchers blinded to experimental conditions, and statistical significance was set at P < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization, Z.L. and L.Z.; Methodology, L.Z., J.W., and Y.L; Validation, Z.L. and L.Z.; Formal analysis, Z.L. and J.W; Investigation, Z.L., J.W., J.C., Y.L., Y.D. H.G., B.Z. and H.O. Resources, J.C., Y.D. and H.G. Data Curation, Z.L. and L.Z.; Writing - Original Draft, Z.L. and L.Z.; Writing – Review & Editing, J.W., Y.L. and K-F.S.; Visualization, Z.L. and L.Z.; Supervision, K-F.S., J.W. and L.Z.; Project Administration, L.Z., J.W. and K-F.S.; Funding Acquisition, L.Z. and K-F.S.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data generated or analyzed during this study are included in this published article (and its supplementary information files). Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-56258-2.