The human motor cortex contributes to gravity compensation to maintain posture and during reaching

Abstract

The neural control of posture and movement is interdependent. During voluntary movement, the neural motor command is executed by the motor cortex through the corticospinal tract and its collaterals and subcortical targets. Here we address the question of whether the control mechanism for the postural adjustments at nonmoving joints is also involved in overcoming gravity at the moving joints. We used single-pulse transcranial magnetic stimulation to measure the corticospinal excitability in humans during postural and reaching tasks. We hypothesized that the corticospinal excitability is proportional to background muscle activity and the gravity-related joint moments during both static postures and reaching movements. To test this hypothesis, we used visual targets in virtual reality to instruct five postures and three movements with or against gravity. We then measured the amplitude and gain of motor evoked potentials in multiple arm and hand muscles at several phases of the reaching motion and during static postures. The stimulation caused motor evoked potentials in all muscles that were proportional to the muscle activity. During both static postures and reaching movements, the muscle activity and the corticospinal contribution to these muscles changed in proportion with the postural moments needed to support the arm against gravity, supporting the hypothesis. Notably, these changes happened not only in antigravity muscles. Altogether, these results provide evidence that the changes in corticospinal excitability cause muscle cocontraction that modulates limb stiffness. This suggests that the motor cortex is involved in producing postural adjustments that support the arm against gravity during posture maintenance and reaching.

NEW & NOTEWORTHY Animal studies suggest that the corticospinal tract and its collaterals are crucial for producing postural adjustments that accompany movement in limbs other than the moving limb. Here we provide evidence for a similar control schema for both arm posture maintenance and gravity compensation during movement of the same limb. The observed interplay between the postural and movement control signals within the corticospinal tract may help explain the underlying neural motor deficits after stroke.

INTRODUCTION

The neural control and the biomechanical tasks of posture and movement are interdependent. When we move our limbs in the gravitational environment, this affects all our body segments, both those that move and those whose posture needs to be maintained. For example, during a reaching movement, the trunk posture is maintained with a neural mechanism termed anticipatory postural adjustment (1–4). A similar mechanism is likely engaged when a wrist joint requires active postural control to prevent the hand from drooping during reaching. Postural adjustments are typically subdivided into those preceding movement and those accompanying movement (5); here we focus on the latter. Extensive evidence in animals shows that the postural adjustments accompanying movement are initiated and shaped by the output of the primary motor cortex via the corticospinal tract through its collaterals to the reticulospinal tract (5, 6). The importance of the reticulospinal tract in the integration between posture and movement, including reaching, is well established in cats (5, 7) and primates (8, 9). A recent study in mice has shown not only that the brain stem circuits support such a control scheme but also that the forelimb musculature involved in gravity compensation, e.g., elbow flexors and wrist and finger extensors, receives diverse inputs from several nuclei in reticular formation that are necessary for performing reaching and grasping tasks (10). Both posture maintenance and postural adjustments during movement serve the same goal: overcoming gravity. Furthermore, they are interdependent, as the amplitude and direction of postural adjustments that accompany movement depend on the initial posture. Therefore, it is logical to suggest that the neural control pathways involved in maintaining posture and in generating the postural adjustments accompanying movement would be the same. To test this idea, we compared the excitability of the same corticospinal pathways involved in producing arm muscle contractions during a static postural task and a reaching task.

The idea that holding still and moving rely on separate, although interdependent, control signals is at the forefront of motor control research (11). The interdependence between the two types of neural control signals has been suggested to result from the integration of the movement commands to compute the appropriate postural command (12). This idea can be thought of as the mathematical equivalent of the anatomical substrate on which the neural signals of the corticospinal tract are being integrated by the reticulospinal tract (5, 7–9). In this study, we too subdivide the reaching movement into posture- and movement-related components, guided by the terms of the equations of motion. The posture-related component of the reaching movement is the transition from the starting posture to a new posture. This component comprises the terms in the equations of motion that contain the gravity acceleration constant. When the equations are solved for the applied torque about a given degree of freedom, this component represents the combined effect of muscle forces that compensate for the gravitational moments that change with the rotating segments during movement. The movement-related component comprises the residual terms in the equations of motion that do not contain the gravity acceleration constant. It encompasses the muscle forces needed for propulsion, i.e., accelerating the limb away from the starting position and decelerating it to stop at the goal position. This accomplishes what can be termed as focal movement, i.e., the translation of the end point between the starting and ending posture irrespective of gravity. The individual muscle contribution to these posture- and movement-related components of reaching can be extracted from the temporal profiles of electromyographic (EMG) signals. The former corresponds to the static component of EMG defined as the linear or cosine ramp (up or down) in the recruitment of multiple motoneuron pools during reaching movement. The latter is the phasic component of EMG we typically observe as bursts (13–15). Here we suggest that the posture-related component of reaching, or the corresponding static EMG, constitutes a postural adjustment accompanying the reaching movement. To test this idea, we hypothesize that the corticospinal excitability is proportional to static EMG and the gravity-related muscle torques during both static postures and reaching movements.

In humans, the excitability of the corticospinal tract and its collaterals can be measured with single-pulse transcranial magnetic stimulation (TMS). TMS over the motor cortex activates the corticospinal tract and its collaterals, which in turn activate spinal motoneurons, causing motor evoked potentials (MEPs) that can be observed in EMG. MEPs provide a measure of instantaneous excitability or recruitment of both pre- and postsynaptic elements of the corticospinal tract and its collaterals (16–19). TMS at low intensity does not disrupt neural state (20); therefore changes in MEP amplitudes can reveal the changes in the descending contribution to muscle recruitment during posture and movement. For example, MEP amplitudes have been shown to change with different hand postures, indicating modulated corticospinal contribution to the recruitment of muscles supporting those postures (21). The amplitude of a given MEP also depends on the excitability of the motoneuron pool of the parent muscle (22–25). This MEP amplitude metric was used to test the primary hypothesis stated above. Furthermore, the proportional contribution of the presynaptic elements to muscle recruitment, termed corticospinal gain, can also be estimated as the proportion between MEP amplitude and the amplitude of the recruitment of motoneuronal pools observed with EMG in absence of MEPs, termed background EMG (26). During movement, the corticospinal gain has been shown to change in a phase-dependent manner independently from the background EMG in wrist muscles and in proximal antigravity muscles (27). In other words, MEP amplitude can go up when the EMG goes down, indicating an increase in the excitability of the stimulated descending pathways and a proportional increase in their contribution to the recruitment of the corresponding motoneuron pool relative to the contribution of the unstimulated neural pathways. Here we directly compare the corticospinal gain in the same muscles involved in both posture and movement. Using this method, we tested a secondary hypothesis that the corticospinal gain is proportional to the gravity-related joint moments during both static postures and reaching movements.

MATERIALS AND METHODS

Subjects

The Institutional Review Board approved all procedures in this study (Protocol No. 1309092800). Potential participants with any musculoskeletal pathologies, injuries, prior history of seizures, fainting, or tinnitus or those taking prescribed psychoactive medications were excluded. We obtained written informed consent before the start of experiments from 10 healthy individuals (6 male, 4 female, 24.3 ± 1.8 yr old, 76.3 ± 14.5 kg). All participants reported to be right-hand dominant.

Specifying and Recording Movement

During the experiment, arm postures and reaching goals were defined in a virtual reality environment created with Vizard software (WorldViz) and an Oculus headset (Fig. 1A). Spherical virtual targets 8 cm in diameter were displayed in the sagittal plane and defined a set of postures and reaching movements as described in detail in Ref. 28 (Fig. 1B). The movements were similar to the planar movements in the horizontal plane in Gritsenko et al. (26) but occurred in a vertical sagittal plane without external mechanical constraints. The locations of virtual targets defined five posture or three reaching movements between them we termed Control, Resistive, and Assistive movements. The terms reflect the impact of interaction torques between the arm segments from the rotations of the shoulder and elbow joints that either counteract the net applied torques (Resistive) or amplify the net applied torques (Assistive), with the third movement having an intermediate effect (Control) as in Gritsenko et al. (26).

Figure 1.

Experimental paradigm. A: participant’s view of the targets in virtual reality. Black lines connect virtual representation of LED markers and show the arm position and orientation in virtual reality relative to targets. The red sphere is the starting target; the green sphere is the reaching goal. The yellow sphere represents the fingertip LED on the index finger, which the participants were instructed to move to the center of the green target. B: schematic of the dynamic model used to calculate joint angles and torques from motion capture and representative end point trajectories from the 3 movements. Three segments (beige ovals) were used to simulate inertial properties of upper arm, forearm, and hand; centers of mass are indicated by black and white circles. Positive directions for each joint [degree of freedom (DOF)] are shown with black circular arrows. The start and stop target locations are shown for the Control, Resistive, and Assistive movements (black, red, and blue, respectively). The motion capture marker locations are shown as yellow circles. C: visual representation of the experimental paradigm. D: example timeline of events during a single reaching movement. GO represents the color cue to start the movement. RMT, resting motor threshold; RT, reaction time; TMS, transcranial magnetic stimulation.

To minimize the intersubject variability in kinematics during experiments, the locations of virtual targets were adjusted based on individual segment lengths. The locations of the virtual targets were calculated with trigonometry based on the desired shoulder and elbow angles and segment lengths of the subject. The targets were displayed relative to each individual’s shoulder marker on the acromion (Fig. 1B). The experiment was subdivided into two sessions, session 1 with static postures and session 2 with reaching movements. In both sessions the participants were asked to reach to or to hold at the virtual target with their right (dominant) arm without moving the trunk while keeping the elbow close to the trunk and a neutral and pronated wrist. The locations of virtual targets were the same for both sessions, so that posture numbers correspond to the start and stop targets for the subject reaches shown in Fig. 1B.

During session 1, participants were asked to reach to a virtual target and hold there for up to 1 min before moving to the next target and holding there. The starting position for the Resistive movement was the same as the ending position for the Control movement; therefore five postures were explored (Fig. 1B). While the participant was holding the arm in one posture, we applied 12 TMS pulses >5 s apart at 90% of resting motor threshold (RMT; see below for details). Six or more TMS pulses has been shown to provide sufficiently reliable estimates of MEPs (29).

During session 2, participants were asked to reach between two virtual targets. Each trial started with the appearance of start (green) and stop (red) targets (Fig. 1D). The colors and target locations did not change until the participants placed their index finger, indicated with a yellow sphere, into the start target. One second after this occurred, the stop target changed color from red to green, directing individuals to begin the movement (Fig. 1D). The target pairs for a given movement were presented to each individual in the same pseudorandom order generated before the study. Each movement was repeated 138 times (total of 414 trials). The movement trials were divided into stimulation trials (126 per movement, 378 total) and nonstimulation control trials (12 per movement, 36 total) interspersed randomly. During the stimulation trials TMS was performed at 90% of the RMT. In one half of trials (189 of 378), the TMS was triggered once per trial at a random delay of 0–550 ms after the participant touched the start target. This triggering method ensured that MEPs were observed either preceding or around the movement onset time. In the other half of stimulation trials, the TMS pulses were triggered once per trial after the participant left the start target at a random delay of 0–550 ms. This triggering method ensured that MEPs were observed through the duration of the movement. The timing of each TMS pulse was recorded for post hoc normalization to the movement phase and binning described below.

Motion capture was performed with the Impulse system (PhaseSpace). Nine light-emitting diode markers were placed on bony landmarks of the arm and trunk according to the best practice guidelines (30). Marker coordinates were sampled at 480 Hz with Recap software (PhaseSpace). The marker locations were also streamed to virtual reality and used to represent the three main segments of the arm (hand, forearm, and upper arm) as a stick figure (Fig. 1A). Motion capture, electromyography (EMG; details below), and virtual events were synchronized with custom hardware as described in Ref. 31.

Procedure for Calculating Joint Torques

After experiments, motion capture data from session 2 were low-pass filtered at a cutoff frequency of 10 Hz. The mean residuals of marker triangulation ± standard deviation across individuals were 5.7 ± 0.45 mm. Joint angles were calculated by defining local coordinate systems of four rigid bodies for the trunk, humerus, forearm, and hand using at least three markers per rigid body (Fig. 1B). To calculate active joint torques, we used a dynamical model as described in Ref. 28. Briefly, the model consisted of three segments (humerus, radius/ulna, and hand). The segment inertia was calculated from cylinders for humerus and radius/ulna segments and from a rectangular prism for the hand. The segments were connected through 3 degrees of freedom (DOFs) at the shoulder joint, 1 at the elbow joint, and 3 at the wrist joint. The model was built in Simulink (MATLAB); it is included in the Supplemental Data. The model was customized to individual morphology by scaling segment length and inertia based on an individual’s height and weight with published average proportions (32). Inverse simulations were run with the scaled models using the angular kinematic trajectories averaged per reaching direction to obtain applied joint torques for a given individual and movement type, termed muscle torques. The muscle torques were verified with forward simulations, and the simulated angular kinematics was compared to the angular kinematics obtained from motion capture data by calculating a root-mean-squared error (RMSE: 0.059 ± 0.035 rad). Positive rotations defined by local coordinate systems are illustrated in Fig. 1B. Analysis was done on the angles and torques around the x-axes, i.e. flexion/extension degrees of freedom at all three joints, because motion was primarily in the sagittal plane and out-of-plane torques were negligible. The postural component of muscle torque, termed postural torque, was extracted by running simulations without gravity and subtracting those from the muscle torques simulated in presence of gravity as described in detail in Ref. 15. The postural torque has been shown to follow closely the static component of EMG during similar reaching movements (15).

Electromyography

Muscle activity and responses to TMS were recorded in 12 upper limb muscles with Trigno (Delsys Inc.), a wireless surface EMG system. The recorded muscles included four muscles spanning the shoulder, three muscles spanning both shoulder and elbow, two muscles spanning only the elbow, and three muscles spanning the wrist (Table 1). Muscles were identified on the basis of anatomical landmarks and palpation during contraction; EMG sensors were placed on muscle bellies oriented longitudinally along the muscle fibers. EMG signals were sampled at 2 kHz with a gain of 1,000. EMG recordings were high-pass filtered at 10 Hz to remove any signal drift and rectified before any EMG/MEP quantification. EMG profiles from trials without TMS were low-pass filtered at a cutoff frequency of 20 Hz and normalized per individual using a maximum value for each muscle across all movement directions.

Table 1.

Pearson correlation coefficients between MEPs and background EMG

Muscle	Abbr	J	Static r	Dynamic r
				Control	Resistive	Assistive
Pectoralis	Pec	S	0.31:0.45:0.59	0.40:0.49:0.74	−0.33:0.63:0.80	0.11:0.49:0.56
Anterior deltoid	AD	S	0.71: 0.8:0.82	−0.08:0.60:0.83	0.51:0.67:0.93	0.19: 0.6:0.85
Posterior deltoid	PD	S	0.34:0.43:0.64	−0.13:0.36:0.62	0.63:0.75:0.92	−0.59:0.24:0.68
Teres major	TM	S	0.52:0.68:0.76	0.30:0.48:0.88	0.64:0.82:0.97	−0.01:0.45:0.80
Triceps (long head)	TriLo	E	0.52:0.67:0.75	−0.27:0.03:0.32	0.46:0.64: 0.80	0.22:0.69:0.73
		S
Triceps (lateral head)	TriLa	E	0.33:0.41:0.64	0.12:0.45:0.77	−0.38:0.34:0.76	0.76:0.81:0.88
Biceps (long head)	BicL	E	−0.09:0.19:0.53	0.37:0.55: 0.80	−0.1:0.21:0.69	0.34:0.55:0.85
		S
Biceps (short head)	BicS	E	0.18:0.28:0.48	0.39:0.74:0.84	0.04:0.32:0.72	0.23:0.65:0.84
Brachioradialis	Br	E	0.03:0.31:0.52	0.37:0.75:0.89	0.62:0.65: 0.70	0.25:0.67:0.74
		W
Flexor carpi radialis	FCR	W	−0.14:−0.08:0.18	−0.35:−0.04:0.60	0.35:0.56:0.76	−0.39:0.39:0.86
Flexor carpi ulnaris	FCU	W	−0.08:0.00:0.27	−0.55:−0.34:0.12	−0.34:0.54:0.65	−0.17:0.71:0.91
Extensor carpi radialis	ECR	W	−0.13:0.32:0.36	−0.4:0.53:0.78	0.16: 0.8:0.87	0.05:0.58:0.77

Values are Pearson correlation coefficients between motor evoked potential (MEP) and electromyography (EMG) (r) shown as 25th percentile:median:75th percentile across individual values. J stands for joints spanned by muscle, which includes the flexion/extension degrees of freedom of shoulder (S), elbow (E), and wrist (W) joints. These were included in the linear regressions for the analysis of the relationships between MEPs and postural torques. Abbr, abbreviated muscle names.

Transcranial Magnetic Stimulation

We assessed the corticospinal contribution to individual muscles with single-pulse TMS delivered by the Super Rapid stimulator with a figure-of-eight coil (Magstim). The coil was oriented tangentially to the scalp at a 45° angle to the midline with the handle pointing posteriorly and laterally (Fig. 1D). The coil location over the scalp and its orientation were maintained with the Brainsight neuronavigation system (Rogue Research). This coil positioning has been shown to preferentially activate pyramidal tract neurons (33–35) and to generate responses proportional to cortical activity.

Focal TMS with the figure-of-eight coil causes MEPs in multiple muscles (36, 37). The amplitudes of these MEPs and which muscles are involved depend on the strength of the magnetic field generated by the TMS device, the location and orientation of the coil over the scalp, and individual anatomy. The location of the coil was selected with the hot-spot method (38, 39), during which the coil was moved over the estimated location of the primary motor cortex until a MEP of at least 50-µV amplitude was evoked in biceps. This method controlled for the anatomical differences between individuals and defined a consistent stimulation location on the motor homunculus (40). This was done with the anticipation that stimulating the same anatomical location with the lowest stimulation threshold measured in biceps would activate the same component of the whole corticospinal tract across individuals and produce similar responses in other muscles across individuals.

The strength of the magnetic field is typically tailored to individuals with a resting motor threshold (RMT) method, i.e., identifying the lowest magnetic field strength that evokes MEPs 50% of the time. This reduces the intersubject variability in TMS responses. RMT was determined at the hot-spot location by varying the stimulation intensity until a MEP > 50 µV was evoked 50% of the time in biceps (long head) (BicL). This procedure ensured that the stimulation amplitude was adjusted to individual differences in corticospinal excitability at rest at the time of experiment and thus further minimized intersubject differences in MEP amplitudes. A single TMS pulse was delivered in a given trial, and each trial lasted 5 s. This was done to minimize the instances when two TMS pulses would be delivered at >0.2 Hz instantaneous frequency to avoid any long-term changes in corticospinal excitability (41). The experiment consisted of two consecutive sessions conducted on the same day. The number of TMS pulses at some phases of reaching movements was limited because of the time constraints of 3-h experiments in an effort to prevent fatigue. The minimum number included in each analysis is 20 MEPs.

The amplitudes of MEPs produced with the described method are stable across days (42, 43) and have sufficient spatial resolution to differentiate proximal and distal muscles (44). Here MEPs were quantified with two methods, peak-to-peak amplitude and area methods commonly used in other studies (39, 45, 46). MEPs in upper limb muscles were observed within the time period lasting from 10 ms to 50 ms after the TMS pulse (Fig. 2A). The peak-to-peak MEP amplitude was calculated by subtracting the minimum from the maximum of the EMG trace within this time period (Fig. 2B). It was used to calculate the probability of MEPs. The MEP area was calculated by first rectifying the EMG trace within this time period and then integrating it (Fig. 2C). All but the probability measures were derived with the MEP area method.

Figure 2.

Quantifying motor evoked potential (MEP) amplitude. A: biceps electromyographic (EMG) trace aligned on transcranial magnetic stimulation (TMS) pulse (red line) during reaching trials from the Control movement in 1 individual. Gray area shows the window used to quantify MEP amplitude and MEP area. BicL, biceps (long head). B: MEP averaged over 11 trials from BicL EMG traces shown in A. EMG was aligned on TMS pulse (red line), and peak-to-peak MEP amplitude was calculated with the shaded area. C: MEP averaged over 11 trials of rectified EMG traces shown in A. EMG was aligned on TMS pulse (red line), and MEP area was calculated by integrating the shaded area under the curve. D: EMG from flexor carpi ulnaris (FCU) muscle in 1 participant performing the Assistive movement. MEP area values were binned according to the phase of movement, where 0 denotes the kinematic onset of movement and 1 denotes the kinematic offset. MEPs (red) were obtained in n = 126 trials with at least 5 MEPs in each bin. Background EMG is from n = 12 trials without TMS. E: circuit diagram for the analysis of MEPs and MEP gain. Triangle represents neural circuits perturbed by TMS at the cortical level of central nervous system (CNS); hexagon represents neural circuits perturbed by TMS at the brain stem level of CNS, which also receive collaterals from the corticospinal tract; and diamond represents spinal motoneurons, the output of which is used to observe MEP. The integration symbol represents the assumption that these circuits integrate their inputs at a variable rate.

For MEPs collected during static postures in session 1, termed static MEPs, both peak-to-peak and area methods were included in the analysis. The background EMG was calculated over the 40-ms time window directly preceding the TMS pulse with the corresponding peak-to-peak or integrated method, respectively. We defined the presence of a MEP as a peak-to-peak amplitude of at least 5 standard deviations above background EMG amplitude. The probability of evoking MEP was calculated by dividing the number of detected MEPs by the number of stimuli under the same conditions. To determine the consistency of MEPs based on MEP area metric, we calculated the coefficient of variation (CV) across repetitions of the same movement per individual per muscle.

The peak-to-peak MEP amplitude (MEP_amp) during isometric contraction has been shown to follow the Boltzmann equation (24, 47, 48):

$$\mathrm{ME}{\mathrm{P}}_{\mathrm{amp}}=\frac{\mathrm{ME}{\mathrm{P}}_{\max }}{1+e^{\left(\frac{S_{50}-S}{k}\right)}}+B\times \mathrm{P}\text{r}\mathrm{eEMG}+C_1$$ (1)

where MEP_max is the maximum MEP amplitude, S₅₀ is the stimulation intensity to elicit a MEP at 50% of the maximum amplitude, S is the stimulation intensity, k is the recruitment curve slope obtained at different stimulation intensities, PreEMG is the amplitude of EMG measured directly preceding stimulation, and C₁ is a constant. In this study, we performed all stimulations at a consistent intensity of 90% of the RMT, which makes the first term a constant. Therefore Eq. 1 can be simplified and adapted for examining the linear relationships between EMG and MEP area as follows:

$$\mathrm{ME}{\mathrm{P}}_{\mathrm{amp}}=B\times \mathrm{bEMG}$$ (2)

where MEP_amp in our experiment is MEP area measured as described above, the bEMG term is the background EMG calculated immediately preceding TMS pulse in session 1, and B is the slope of the linear relationship between MEP and EMG during static postures. The offset was assumed to be equal to 0, i.e., regression goes through zero with no offset, expecting no MEPs in fully relaxed muscles because of stimulation below RMT. Because the background EMG in some non-load-bearing muscles varied across postures much less than during movement, the most reliable estimate of B for a given muscle was a simple ratio between the median MEP and the median background EMG. To check this assumption, we repeated all analyses using B estimated either as a slope or as a ratio with equivalent results. This article only includes data with static B calculated as a ratio. We derived B in static session 1 to estimate the baseline level of corticospinal excitability during static postures and to compare it to the levels of corticospinal excitability during movement.

The coefficient B is the estimate of the overall excitability of the neural circuits comprising the corticospinal tract and its collaterals relative to the underlying excitability of the motoneuron pool of a given muscle (Fig. 2E). If we assume that the TMS pulse perturbs the neural circuitry by the same amount when it is applied at the same amplitude of the magnetic field over the same cortical location, then the descending contribution to the recruitment of a given muscle is proportional to the excitability of the motor cortical output layer cells and their collaterals, simplified here to only the indirect parallel pathway from the brain stem (Fig. 2E). Therefore, the coefficient B represents a lumped estimate of the excitability of the descending, both direct and indirect, contribution to the recruitment of a given muscle and the excitability of the spinal motoneuron pool of the corresponding muscle. If the excitability of these circuits increases, larger MEPs will be observed at the same levels of background EMG, leading to larger B values, and vice versa.

We calculated MEP latencies with a procedure similar to that used to determine kinematic onsets. Unrectified EMG signals were aligned on the TMS pulse and averaged across repetitions of the same postures. The MEP latency was defined as the time from the TMS pulse to the first maximum of the third derivative of this averaged trace.

For MEPs collected in session 2, termed dynamic MEPs, both peak-to-peak and area methods were included in the analysis. Most data shown are based on MEP area. Outcomes of all analyses applied to both peak-to-peak and area values were equivalent, although the former was more variable than the latter. All trials during which TMS was applied were included in the MEP metrics without determining directly whether a MEP was evoked or not in a given trial. This is because of the difficulties in distinguishing MEPs from compound motor unit action potentials in the presence of changing EMG during movement. MEP area and the CV values were calculated as in session 1. The MEP area values obtained from single trials were grouped into five temporal bins; each corresponded to 20% increments of phase duration from onset to offset of movement (Fig. 2D). This binning procedure grouped MEPs occurring at similar times during movement and provided adequate repetitions to estimate median values. Additionally, TMS responses that occurred up to 20% phase duration before movement were grouped into a bin 0 (Fig. 2D). We defined a minimum repetition criterion of 5 MEPs based upon Ref. 29. Bins that contained <5 MEPs were excluded from subsequent analyses.

To estimate the changes in corticospinal excitability during movement, we calculated MEP gain as B in Eq. 2. The background EMG was sampled from trials without TMS at the same phase as TMS pulses in the other trials of session 2. We selected short chunks of EMG at the corresponding phase of movement as MEPs and processed them the same way as the MEP area or peak-to-peak metric (Fig. 2, B and C). The selection and binning of background EMG to match the phase of MEPs controls for the variable excitability of the spinal motoneuron pools during movement, thus controlling for the rate of integration of the spinal motoneuron pools (Fig. 2E). The binning procedure ensured the same limb state (limb position and velocity) and thus the same muscle recruitment levels in trials in which the background EMG and dynamic MEPs were obtained.

Results in text are reported as mean values ± standard deviation across individuals, unless otherwise indicated. MEPs are calculated based on the area method unless otherwise indicated. Static MEP gain values in all figures were derived from ratios as shown as green lines in Fig. 5. Dynamic MEP gain values in all figures were derived from regressions between binned MEPs and binned background EMG as shown by regressions in Fig. 6.

Figure 5.

Static motor evoked potential (MEP)/electromyography (EMG) regressions. Linear regressions between static MEPs and background EMG per muscle are shown for a single participant. Black symbols represent individual MEPs collected during each posture. The black line denotes the regression fit, and the green lines denote the mean MEP and mean EMG used to calculate the ratio. AD, anterior deltoid; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; ECR, extensor carpi radialis; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

Figure 6.

Dynamic motor evoked potential (MEP)/electromyography (EMG) regressions. Linear regressions between dynamic MEPs and EMG are shown per muscle for the same participant as in Fig. 5 for each movement (Control, black; Resistive, red; Assistive, blue). Dots represent the mean MEP/EMG value at each time bin during the movement, and the horizontal and vertical diameters of ovals denote the SD of EMG and MEP values, respectively. Green lines show mean static MEPs and mean background EMGs from Fig. 5. AD, anterior deltoid; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; ECR, extensor carpi radialis; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

Statistical Analyses

The linear relationships between MEPs and the background EMG were quantified with the Pearson correlation coefficient (r) and MEP gain (ratio B in Eq. 2) across all postures for static MEPs or per movement for dynamic MEPs (slope B in Eq. 2) for each muscle and individual separately. For static MEPs, the background EMG was a section of preceding EMG from the same trial processed the same way as MEPs. For dynamic MEPs, the background EMG was a section of EMG processed the same way as MEPs from trials without TMS at the corresponding phase of movement. For the static MEPs from session 1, the linear regressions were fitted between single-trial MEPs and corresponding preceding EMG across all five postures shown in Fig. 1B. For the dynamic MEPs from session 2, the linear regressions were fitted between median MEPs and median EMG across movement phases (bins 0–5; e.g., between recorded MEP and background EMG in Fig. 2D). The r and MEP gain values were compared across conditions (Static, Control, Resistive, and Assistive) with the nonparametric Kruskal–Willis tests, using the kruskalwallis function in MATLAB. Two separate tests were done to compare the 10 × 48 array of correlation coefficients for each of 10 subjects grouped by muscle type (12 muscles) or grouped by conditions (Static, Control, Resistive, and Assistive). The chi-square statistics and the probability of making the type I error (P values) are reported in results with the number of degrees of freedom in parentheses, e.g., χ²(11) and χ²(3), for comparisons across muscles and across conditions, respectively. Another two tests were done to compare the MEP gain values of the same relationships. The post hoc comparisons were done with the multicompare function in MATLAB.

To test the hypothesis, we regressed all EMG-based metrics against the postural torque in all conditions. We expect that the intersubject variability in EMG values may contain biomechanically relevant information, such as the changes in postural torque amplitudes due to subject size, in addition to the changes in postural torques across conditions. Therefore, we concentrated our analysis on the changes in torques between conditions in individuals. We standardized the data by demeaning and normalizing to the range so that across-participant comparisons could be made. First, we calculated individual differences in postural torques between postures using data from session 1. These values corresponded to the start and stop locations of each movement in session 2; therefore the difference represents the change in postural torque between starting or between stopping positions across movements. Specifically, the values of postural torques in session 1 from the two hold phases in Fig. 3B were subtracted from each other resulting in 6 values per DOF per individual and analyzed together with n = 60 across individuals per DOF. Next, the values of postural torques in session 2 from each of bins 0–5 in Fig. 3B were subtracted from their corresponding values in the other movements. This resulted in 6 demeaned values of postural torques between movements for each individual and each DOF. Finally, we then normalized the demeaned postural torque values by dividing all values by their range (maximum − minimum) across individuals and across movements or postures for each DOF separately. These dimensionless postural torques still captured the intersubject variability, so that subjects whose postural torques varied less between postures or movements had lower torque values than those whose postural torques varied more between postures or movements. Unitless torque values from the acceleration phase (bins 0, 1, and 2, n = 30 across individuals per DOF) and deceleration phase (bins 3, 4, and 5, n = 30 across individuals per DOF) were analyzed separately. This was done because MEP amplitudes changed throughout the movement.

Figure 3.

Movement kinematics and dynamics. A: joint angles calculated from motion capture. B: the component of muscle torque that compensates for gravity load on the joints, termed postural torque.

The EMG-based metrics such as background EMG, MEPs, and MEP gain values from all conditions were processed the same way as the postural torques. For example, the median MEP gain in flexor carpi ulnaris (FCU) at bin 0 in Control movement (Fig. 2D) was subtracted from the corresponding value in Assistive movement. To ensure that each individual contributed equally to the regressions, the demeaned EMG-based metrics were scaled to the maximal values across all movements or postures per individual. The dynamic MEP gain values were not scaled this way, as they were already normalized to individual static MEPs using the slope described in Eq. 2. Only data in which MEP amplitude was calculated using the area method shown in Fig. 2C are included in this article.

The unitless values of EMG-based metrics were regressed against the corresponding unitless values of postural or muscle torque. The EMG-based metrics in a given muscle was matched to the postural or muscle torque metric from the DOF that spans the corresponding joint (Table 1). For BicL and triceps (long head) (TriLo), only the EMG and MEP regressions against the postural or muscle torques at the shoulder are included in this article. Matching BicL and TriLo with elbow DOF led to the same conclusions; these regressions were similar to those between biceps (short head) (BicS) and lateral head of triceps (TriLa) and postural or muscle torque at the elbow (data not shown). The statistical significance of linear dependencies between EMG metrics from 12 muscles and the postural or muscle torque of the corresponding joint was evaluated with the repeated-measures analysis of variance (manova function in MATLAB). Fifteen tests were done: 1) EMG vs. postural torque in session 1; 2) static MEP amplitude vs. postural torque in session 1; 3) static MEP gain vs. postural torque in session 1; 4) EMG vs. postural torque in acceleration phase; 5) dynamic MEP amplitude vs. postural torque in acceleration phase; 6) dynamic MEP gain vs. postural torque in acceleration phase; 7) EMG vs. postural torque in deceleration phase; 8) dynamic MEP amplitude vs. postural torque in deceleration phase; 9) dynamic MEP gain vs. postural torque in deceleration phase; 10) EMG vs. muscle torque in acceleration phase; 11) dynamic MEP amplitude vs. muscle torque in acceleration phase; 12) dynamic MEP gain vs. muscle torque in acceleration phase; 13) EMG vs. muscle torque in deceleration phase; 14) dynamic MEP amplitude vs. muscle torque in deceleration phase; 15) dynamic MEP gain vs. muscle torque in deceleration phase. The number of samples for regressions in tests 1, 2, and 3 was n = 720 comprising 6 values for 12 muscles in 10 individuals and an equal number of matching torques. The number of samples for the rest of regressions was n = 360 comprising 3 values per phase for 12 muscles in 10 individuals and an equal number of matching torques. The post hoc analysis of the goodness of fit of linear regressions was done with the regress function in MATLAB. Familywise error for multiple regressions across 12 muscles was corrected with the Sidak method: α = 1 − (1 − 0.05)^1/12 = 0.0043.

RESULTS

The virtual targets were effective in evoking the desired behavior and standardizing the movement kinematics and dynamics across all individuals (Fig. 3). Holding the arm against gravity in the five postures and transitioning between them during reaching was accomplished with postural torques of different magnitudes at the major arm joints (Fig. 3B). The postural torques were larger during static postures in session 1 (Fig. 3B, hold) than at the start and end of movements in session 2 (Fig. 3B, bins 0 and 5) because of the additional contribution of the propulsive torques to accelerate and decelerate the arm during movement. However, the directions of changes of postural torques across movements at matching postures were the same in both sessions. For example, the postural torque about the shoulder in the starting position for the Resistive movement and at posture 2 were both smaller than that at the starting position for the Control and Assistive movements and at postures 1 and 4 across the two sessions (Fig. 3B, shoulder). The shoulder motion in the Control and Assistive movements was with gravity; both were accompanied by decreasing postural torque at the shoulder, whereas in the Resistive movement the shoulder motion was against gravity and the postural torque was increasing (Fig. 3B, shoulder). In contrast, the postural torques about the elbow and wrist joints were more alike in the Assistive and Resistive (Fig. 3B, elbow and wrist, blue and red) movements despite different motion of the elbow in these two movements (Fig. 3A, elbow and wrist, blue and red). The postural torques at the elbow and wrist joints were equal across all starting postures and in the beginning of movement; they were both diminishing during reaching in the Assistive and Resistive movements but not in the Control movement. These results suggest that the postural control of the elbow and wrist when the wrist is not moving can be coupled, because the gravity-related postural torques at the elbow and wrist are changing together. In contrast, the postural control of the shoulder needs to be decoupled from that of the elbow and wrist because the gravity-related postural torques at the shoulder are changing in a different direction from that at the elbow. This applies to both the posture-related component of focal movement and maintaining static posture.

During static postures in session 1 static MEPs were observed in most muscles in most subjects at stimulation amplitudes below RMT. For example, biceps muscle participates directly in posture maintenance by holding the elbow and shoulder flexed in the tested postures (Fig. 4A). As expected, we found that the probability of evoking a MEP in this muscle was higher than expected from stimulation at the RMT in most individuals (Fig. 4B). Furthermore, the probability of evoking a static MEP was higher than expected in all muscles, even those not directly involved in supporting the arm against gravity (Fig. 4C). Altogether this shows that the overall corticospinal excitability was increased when the arm was held against gravity in different postures compared with when the arm was relaxed.

Figure 4.

Static motor evoked potentials (MEPs). A: electromyographic (EMG) traces of the biceps muscle during static postures are displayed for all participants. The black line corresponds to the time of transcranial magnetic stimulation (TMS) pulse. B: each bar corresponds to the probability of a MEP occurring for each participant across all postures. The solid black line denotes the probability of MEP occurrence at resting motor threshold (RMT). The dashed line corresponds to the increase in probability that could be detected given the statistical power of our study design (β = 0.9). C: the probability of a MEP occurring in each muscle across participants. The dots represent mean probabilities across participants, and the error bars denote SD across 10 participants. AD, anterior deltoid; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; ECR, extensor carpi radialis; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head). D: MEP latencies for each muscle across participants. The dots represent mean probabilities across participants, and the error bars denote SD across 10 participants. E: the coefficient of variation was calculated for MEPs collected under the same conditions [participant, muscle, and movement (pink) or posture (green)]. Previously reported coefficients of variation (39, 49–51) are shown in black. The boxes denote interquartile range, and the center line denotes median values.

The static MEPs were evoked at similar latencies and with consistent amplitudes across individuals. The average latency across muscles was 15 ± 1.9 ms (Fig. 4D). Dynamic MEP latencies were on average 14.2 ± 2.5 ms, similar to the latencies of static MEPs, indicating that the same pathways were likely evoked by TMS in both sessions. These latencies broadly reflected the proximal to distal distribution of muscles and the associated differences in the distance traveled by the action potentials evoked by TMS.

The MEP amplitudes were highly consistent, as evidenced by the CV of static MEPs equal to 0.35 and dynamic MEPs equal to 0.5 (Fig. 4E), which is comparable to reported values recorded at rest and in relaxed muscles (39, 49–51). The larger CV for the dynamic MEPs compared with the static MEPs is likely driven by the changing motoneuronal excitability reflected in the changing background EMG and muscle lengths during movement (46, 52, 53).

The levels of the corticospinal contribution to the recruitment of individual muscles, termed corticospinal excitability, were estimated with linear regressions between MEPs and background EMG. The static MEPs were linearly related to the background EMG across different postures with variable strengths of relationships across muscles [Fig. 5 shows an example for 1 subject; Supplemental Figs. S1–S5 (available at https://doi.org/10.6084/m9.figshare.21383796.v1) show data for 5 other subjects; Table 1 shows Pearson correlation coefficients between MEPs and EMG across all subjects]. Weaker static MEP-EMG relationships were observed in the distal muscles, likely because of the small changes in EMG of those muscles across different postures, which limited the range of TMS responses. To abstract our conclusions from the reliability of these linear relationships, we report below the static MEP gain calculated based on the ratios between MEPs and background EMG instead of slopes (Fig. 5, green lines). Equivalent results were obtained using the static MEP gain values based on regressions (data not shown).

The amplitudes of dynamic MEPs were also linearly related to the background EMG throughout the movement [Fig. 6 shows an example for the same subject as in Fig. 5; Supplemental Figs. S6–S10 (available at https://doi.org/10.6084/m9.figshare.21383781.v1) show data for 5 other subjects]. The correlation coefficients between dynamic MEPs and background EMG summarized in Table 1 were different between all conditions [χ²[3) = 13, P = 0.0041], with post hoc tests showing significantly higher correlation coefficients in both Resistive and Assistive movements than during static postures (significant: Resistive–Static interval = 53, P = 0.0157 and Assistive–Static = 48, P = 0.0394; not significant: Static–Control = −10, P = 0.9445, Resistive–Assistive = −43, P = 0.0745). The correlation coefficients were different between muscles [χ²(11) = 30.47, P = 0.0013], with post hoc tests showing that the largest change in correlation coefficients was between deltoids and wrist flexor muscles [anterior deltoid (AD)–flexor carpi radialis (FCR) interval = 117, P = 0.0088; AD–FCU = 121, P = 0.0054; posterior deltoid (PD)–FCR = 102, P = 0.0462; PD–FCU = 106, P = 0.0306; correlation coefficients were not different between other muscles]. Overall, these data show that MEPs were proportional to EMG during movement. Furthermore, the MEP gain was not different across tasks across all participants [Fig. 7; χ²(3) = 7, P = 0.0833]. However, the MEP gain was different in different muscles [χ²(11) = 35, P = 0.0003], with post hoc tests showing the largest MEP gain in the lateral head of triceps (TriLa) and brachioradialis (Br) [pectoralis (Pec)–TriLa interval = −83, P = 0.0282; TriLa–FCR = 110, P = 0.0186; AD–Br = −114, P = 0.0133; Br–FCR = 117, P = 0.0085; Br–FCU = 103, P = 0.0431; MEP gain was not different between other muscles]. The differences in MEP gain across muscles are likely due to the different strengths of functional connections to these muscles of the stimulated location of the motor cortex. The differences in MEP gain in distal muscles may also have been driven by the increased strength of correlations between dynamic MEPs and EMG during movement compared to that during static postures. This is because of the different ranges of background EMG during posture maintenance and during movement. The next set of analyses focuses on whether some of the observed differences and variability in corticospinal excitability and its relationship to background EMG reflect the postural forces needed to be exerted by muscles to support the limb against gravity during static postures and during movement.

Figure 7.

Slopes of the motor evoked potential (MEP)/electromyography (EMG) regressions shown in Figs. 5 and 6. The dots denote median values, and the whiskers denote interquartile range across participants. AD, anterior deltoid; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; ECR, extensor carpi radialis; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

During posture maintenance, the forces produced by muscle contractions sum up to produce active postural torques around joint degrees of freedom to support the arm against gravity. Therefore, the EMG changes between postures recorded in our session 1 should be linearly related to the changes in postural torques between postures. When kinematics is controlled for, the strength of this linear relationship is dependent on the size of individuals, as larger limbs need larger postural torques to support them against gravity. To take this meaningful intersubject variability into account, we regressed background EMG levels at different postures and different phases of movement against the corresponding postural torques across individuals. As expected, our data showed that the EMG was linearly related to the individual postural torques across postures with significant slope and intercept [F(11,48) = 1 and 67, P = 0.399 and < 0.001 for intercept and slope, respectively; Fig. 8A]. All monoarticular shoulder muscles, both flexors and extensors, and both triceps and biceps muscles increased their activity with postural flexion torque at the shoulder and decreased their activity with postural extension torque at the shoulder (Fig. 8A, top two rows). The activity of both triceps and biceps muscles also increased with the postural extension torque at the elbow (Fig. 8A, third row). These data show that postural muscle activity changed together in both flexors and extensors, indicating the changing levels of cocontraction that accompany different postures.

Figure 8.

Electromyography (EMG) and postural torque (τ_p) regressions. Linear regressions between EMG and postural torques are shown for the static postures (A) and the acceleration (B) and deceleration (C) phase of movement. Each row denotes a separate joint and includes muscles that span that joint. The shoulder joint is further separated into monoarticular (1st row) and biarticular (2nd row) muscles. Bolded lines highlight statistically significant correlations (P < 0.0043). AD, anterior deltoid; au, arbitrary units; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; DF, dorsiflexion; E, extension; ECR, extensor carpi radialis; F, flexion; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; PF, palmar flexion; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

During reaching, the arm moved from the starting posture to the target posture. This motion was accompanied by changing postural torques produced by muscles to support the arm against gravity. During the acceleration phase of movement, the relationships between EMGs of most muscles and postural torques diminished, although the overall slope was still significant [F(11,88) = 2 and 9, P = 0.140 and < 0.001 for intercept and slope, respectively]. In the deceleration phase of movement, the EMGs were again strongly linearly dependent on postural torques [F(11,88) = 2 and 7, P = 0.073 and < 0.001 for intercept and slope, respectively]. The relationships between EMGs and muscle torques were similar to those for postural torques, with more variance captured by the intercept [acceleration: F(11) = 5, P = 0.001; deceleration: F(11) = 4, P = 0.008] but less variance captured by the slope [acceleration: F(88) = 3, P = 0.018; deceleration: F(88) = 3, P = 0.011]. This shows that both muscle torques and their postural components capture the linear dependencies of moments produced by muscle contractions. These data are consistent with results and conclusions from our earlier study showing that the temporal profiles of active postural torques during reaching can be used to isolate the tonic components of EMG to quantify the muscle recruitment responsible for supporting the arm against gravity during movement (15).

The post hoc analysis of the relationships in individual muscles revealed more nuance. The relationships between shoulder muscles and shoulder torque had the same sign across static postures and both phases of movement (Fig. 8, top row). However, the signs of the EMG-torque relationships between biceps and triceps muscles and postural torques changed in the deceleration phase of movement compared with those in static postures (Fig. 8C). Altogether, these data show that the muscle activity changed with postural torques at all joints in both flexors and extensors. This indicates that changing levels of cocontraction accompany movement phases with different dynamics.

The static MEP amplitudes were also proportional to the postural torques across postures [F(11,48) = 2 and 17, P = 0.149 and < 0.001 for intercept and slope, respectively]. Similarly, there was a strong linear relationship between the dynamic MEPs and the postural torques [acceleration: F(11,18) = 2 and 15, P = 0.083 and < 0.001 for intercept and slope, respectively; deceleration: F(11,18) = 6 and 27, P = 0.001 and < 0.001 for intercept and slope, respectively]. These relationships closely mimicked the EMG-torque relationships in Fig. 8. The MEP linear fit was worse with muscle torques, evidenced by lower F values for slope [acceleration: F(11,18) = 2 and 3, P = 0.129 and = 0.009; deceleration: F(11,18) = 5 and 1, P = 0.002 and = 0.857 for intercept and slope, respectively]. This shows that during both phases of movement the postural torques represent the dynamics of the changing MEPs better than the muscle torques do. Overall, these data support the primary hypothesis, showing that during posture maintenance and during movement the corticospinal excitability is increased in proportion to the forces required to hold the arm against gravity. The changes in the corticospinal excitability largely paralleled the changed in postural muscle activity.

Post hoc analysis also revealed similar changes in sign of the MEP-postural torque relationships in individual muscles (Fig. 9, thick lines). Moreover, the shoulder flexors (Pec and AD) also changed the sign of their relationships with shoulder postural torque in the acceleration phase (Fig. 9, top row). Overall, there are more significant relationships in individual muscles, underlining the stronger main effect. MEPs were proportional to postural torques in both flexors and extensors. This further suggests that during movement the corticospinal tract modulates the recruitment of individual muscles to dynamically create the different cocontraction patterns necessary to overcome gravity-related load on the arm.

Figure 9.

Motor evoked potential (MEP) magnitude and postural torque (τ_p) regressions. Linear regressions between MEPs and postural torques are shown for the static postures (A) and the acceleration (B) and deceleration (C) phase of movement. Each row denotes a separate joint and includes muscles that span that joint. The shoulder joint is further separated into monoarticular (1st row) and biarticular (2nd row) muscles. Bolded lines highlight statistically significant correlations (P < 0.0043). AD, anterior deltoid; au, arbitrary units; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; DF, dorsiflexion; E, extension; ECR, extensor carpi radialis; F, flexion; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; PF, palmar flexion; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

To estimate the changes in the corticospinal excitability that were independent of the linear relationships between MEPs and background EMG, we calculated MEP gain as described in materials and methods. The static MEP gains were not constant across the five postures and varied with postural torques [F(11,48) = 1 and 5, P = 0.781 and < 0.001 for intercept and slope, respectively]. During movement, the dynamic MEP gains were also linearly related to postural torques [acceleration: F(11,18) = 2 and 12, P = 0.056 and < 0.001; deceleration: F(11,18) = 2 and 9, P = 0.063 and < 0.001 for intercept and slope, respectively]. Again, the relationships between dynamic MEP gains and muscle torques were much weaker in both phases of movement [acceleration: F(11,18) = 1 and 4, P = 0.237 and 0.003; deceleration: F(11,18) = 2 and 3, P = 0.210 and = 0.024 for intercept and slope, respectively]. This further shows that during both phases of movement the postural torques represent the second-order dynamics of the changing MEPs better than the muscle torques do.

Post hoc analysis also revealed that the MEP gain-postural torque relationships were changing signs (Fig. 10, thick lines), similar to regressions between MEPs in individual muscles and postural torques (Fig. 9). Additionally, MEP gain in flexor carpi ulnaris was shown to be strongly dependent on wrist postural torque in deceleration phase (Fig. 10C). This shows that not only were MEPs proportional to postural torques but their coefficient of proportionality was also changing in proportion with postural torques. Overall, these data support the secondary hypothesis. This shows that, during both static postures and reaching movements, the strength of the descending contribution to specific muscles changed in proportion with the postural torques needed to support the arm against gravity.

Figure 10.

Motor evoked potential (MEP) gain and postural torque (τ_p) regressions. Linear regressions between MEP gain and postural torques are shown for the static postures (A) and the acceleration (B) and deceleration (C) phase of movement. Each row denotes a separate joint and includes muscles that span that joint. The shoulder joint is further separated into monoarticular (1st row) and biarticular (2nd row) muscles. Bolded lines highlight statistically significant correlations (P < 0.0043). AD, anterior deltoid; au, arbitrary units; BicL, biceps (long head); BicS, biceps (short head); Br, brachioradialis; DF, dorsiflexion; E, extension; ECR, extensor carpi radialis; F, flexion; FCR, flexor carpi radialis; FCU, flexor carpi ulnaris; PD, posterior deltoid; Pec, pectoralis; PF, palmar flexion; TM, teres major; TriLa, triceps (lateral head); TriLo, triceps (long head).

DISCUSSION

Here we used single-pulse TMS to probe corticospinal excitability during unconstrained postural and reaching tasks. We observed that stimulation below RMT evoked MEPs in all muscles when holding the arm in static postures against gravity and during reaching. This means that the corticospinal excitability was increased during posture maintenance and during movement compared to rest, consistent with previous studies (54). Moreover, we have shown that MEPs are linearly dependent on the background EMG (Table 1). This is consistent with previous work showing that a large share of MEP variance can be captured by a simple linear relationship between MEP amplitude and EMG during isometric voluntary contractions (24). Additionally, the MEP gain in distal muscles tended to be higher than in proximal muscles (Table 1, Fig. 7). This is consistent with prior work showing a greater involvement of the corticospinal tract in the voluntary activation of distal versus proximal muscles (55, 56). Such MEP/EMG regressions capture the input-output relationships between the synaptic input into a given motoneuron pool and its output in terms of muscle recruitment measured with EMG (Fig. 2E). This means that when the synaptic input is constant, e.g., evoked by a TMS pulse of a fixed amplitude, the output or the MEP amplitude will vary in proportion to the level of recruitment of the motoneuron pool with the coefficient of proportionality equal to the slope of the linear relationship. In the present study we also observed that the MEP amplitude varied not only with EMG but also with postural torques (Fig. 9). We also showed that EMG varied with postural torques (Fig. 8). Altogether, these results suggest that the changes in corticospinal excitability during posture maintenance contribute to the changes in motoneuron pool recruitment, which cause muscle contractions that, in turn, produce forces needed to support the arm against gravity. These are causal relationships, because they were derived from the stimulation of the motor cortex that drove the responses in EMG, which varied with postural torques. This provides evidence that the motor cortex is involved in producing the forces necessary to support the arm against gravity during posture maintenance.

The changes in MEP gain are also significant. They happen when MEPs change amplitude without the corresponding change in the background EMG or with a different coefficient of proportionality to the EMG (Fig. 2E). These MEP gain changes must be driven by the changes in the synaptic inputs to the motoneuron pools, all things being equal at the spinal level. To keep things equal at the spinal level, we carefully controlled for the intersubject variability in movement kinematics and phase-matched background EMG. This ensured that the underlying recruitment of spinal motoneuron pools had low intersubject variability and could be carefully measured at the moment of arrival of TMS-evoked synaptic input. These careful controls have allowed us to measure the changes in MEP gain. We observed higher MEP gain during movements with complex dynamics (Assistive and Resistive) compared with during static postures (Table 1). We also observed stronger relationships between MEPs and postural torques than between EMG and postural torques in the acceleration phase of reaching (Fig. 8B, Fig. 9B). We have also observed that the MEP gains also changed in proportion to postural torques both during posture maintenance and during movement (Fig. 10). Overall, this provides further supporting evidence that the motor cortex is directly involved in producing the forces necessary to support the arm against gravity not only during posture maintenance but also during movement. It is well established that the neural motor system includes multiple feedback pathways forming a complex sensorimotor control system with both feedforward and feedback elements and internal predictions of limb state (57). Therefore, our data shed light primarily on the feedforward gravity-related modulation of the output of the motor cortex via the corticospinal tract, its collaterals, and downstream targets. However, the observed weaker MEP gain vs. postural torque relationship compared to the MEP vs. torque relationship does suggest that the gravity-related motoneuronal recruitment is likely influenced by spinal feedback mechanisms that modulate the recruitment of the motoneuronal pools in parallel with the descending contribution.

The EMG of antagonistic muscles spanning the same joints, not only the traditional “antigravity” flexors, was proportional to postural torques with the same slope sign across antagonists (Fig. 8). This indicates that cocontraction or mechanical impedance changes in proportion to the forces needed to support the arm against gravity. We have further shown that MEPs and MEP gain in these muscles were also proportional to postural torques without reversing sign across antagonists (Fig. 9, Fig. 10). This indicates that muscle cocontraction is driven by the corresponding changes in corticospinal excitability. The recruitment of multiple muscles to compensate for postural forces further supports the potential involvement of the reticulospinal pathway via the corticospinal collaterals in controlling the mechanical impedance of the limb (58–60). A recent animal study has reported complementary results showing that the reticular nuclei have robust and diverse connections to the forelimb motoneurons, particularly to the elbow flexors and wrist and finger extensors needed to support the forelimb against gravity (10). This is also supported by observations that in people with cortical damage due to stroke the postural muscle torques are less affected than focal movement-related muscle torques (61). Other studies have implicated the recruitment of reticulospinal pathways in the appearance of poststroke gross muscle synergies with characteristic abnormal cocontraction or spasticity (62, 63). Altogether, these observations further support the idea that the corticospinal excitability changes in proportion to postural forces during movement can be driven by the same neural circuitry as the postural adjustments that accompany focal movement.

As defined in introduction, the forces needed for propulsion summate with postural forces, i.e., those needed to support the arm during the transition between the starting posture and the goal posture. These propulsive forces (or torques) reflect complex intralimb dynamics that includes interaction torques; they often change their sign between the acceleration and deceleration phases of movement (64). Our results have shown that in the pectoralis and anterior deltoid the signs of MEP-shoulder torque relationships changed in the acceleration phase compared with the static task and the deceleration phase of movement (Fig. 9, top row). In our earlier study we also found that the sign of the MEP-elbow torque relationship changes during similar reaching movements in horizontal plane (26). We found that the sign of the MEP-torque relationship matches the change in sign of the interaction torques between shoulder and elbow during movements in different directions. Overall, these results suggest that the change in sign of the MEP-torque relationship may reflect the corticospinal involvement in the compensation for intralimb dynamics. The changes in sign of the MEP-shoulder torque relationships that are different from the MEP-elbow relationships are also consistent with the neural strategy of using the shoulder as a leading joint to generate propulsion forces (65).

Although the MEPs and EMGs across recorded muscles were strongly linearly dependent on postural components of muscle torques at shoulder and elbow joints, these relationships for individual muscles were less consistent, as shown by post hoc analysis. Not surprisingly, the weaker relationships were observed by and large when one or both of the variables were not changing or were changing in a very small range. A case in point is the EMG of wrist muscles, which changed very little at different postures because the weight of the hand is low and small changes in EMG relative to noise are harder to detect. Moreover, the range of wrist postural torque in the acceleration phase was lower than in other phases (Fig. 3B, see abscissa ranges in Figs. 8–10). Therefore, the EMG and MEPs in wrist muscles were not related to the wrist postural torques during static postures or during the acceleration phase of movement. In contrast, when the EMG and torque changes were larger in the deceleration phase of movement, relationships between EMG, MEP, and MEP gain appeared. Moreover, despite the absence of the relationships between EMG and MEPs in FCU and wrist torque in the deceleration phase of movement, the relationships between MEP gain and torque were present (Fig. 10C), indicating strong corticospinal modulation of this muscle. In another example, the range of the elbow postural torque in the acceleration phase was lower compared with the static or deceleration phases (Fig. 3B, see abscissa ranges in Figs. 8–10). Consequently, there were no relationships between the EMG of muscles spanning the elbow and the constant elbow torque in the acceleration phase of movement (Fig. 8B). However, the significant relationships did appear for MEP and MEP gain in triceps and elbow postural torque, indicating strong corticospinal modulation of these muscles. Altogether, this shows that the varying strengths of the relationships between the EMG-based metrics in individual muscles and torques reflect biomechanics and strengthen the overall conclusions of the study.

The limitations of study methodology may limit the generality of our conclusions. The first limitation is that the contribution of descending projections from premotor and supplementary motor cortices to motoneuronal excitability cannot be ruled out. In our study we used focal TMS that was targeted and scaled based on responses observed in biceps EMG. This method is designed to limit the volume of current density that is sufficient to stimulate cells located in the arm area of the primary motor and possibly primary sensory motor cortices (44, 66). Therefore, the descending projections from premotor areas were likely not stimulated and their contribution to muscle recruitment and the compensation for passive torques not quantified in the present study. The second limitation is the possibility that using a different muscle for determining the RMT would alter the modulation of MEP gain in other muscles. It seems unlikely for proximal muscles, because the spatial resolution of TMS is rather low. Stimulation over scalp locations 1–2 cm apart has been shown to crudely distinguish between proximal or distal muscle groups (44). Here we have targeted a proximal muscle group by determining our hot spot and RMT based on MEPs in biceps. If we choose another proximal muscle for thresholding, the hot-spot method will likely zero in on the same scalp location but a different magnetic field strength. Increasing or decreasing the magnetic field strength over the same spatial location to target another proximal muscle is likely to activate more or less, respectively, all the same muscles recorded here. The underlying cortical activity that drives the modulation of MEP gain would be the same. However, if we choose a distal muscle for thresholding, the hot-spot method will likely zero in on a different scalp location and a different magnetic field strength. In this case, proximal muscles are not likely to have MEPs because of their generally higher threshold (55, 56). Therefore, the activity of the different underlying cortical area may cause different MEP amplitudes and gains across all muscles compared with those caused by the scalp area chosen based on the biceps hot spot. Consequently, our results generalize to represent a subset of cortical contribution to the control of reaching, but they may not generalize to the more dexterous corticospinal control of distal muscles. The third limitation is that the MEPs produced by TMS during movement may saturate in some muscles at some phases of movement. This possibility is based on observations of similar or lower MEP amplitudes despite increasing background EMG from 10% to 40% of maximal voluntary effort for the former or 25% to >50% of maximal voluntary effort for the latter (48, 67). Here, we did observe linear relationships between background EMG and MEP amplitude during both postural and reaching tasks (Table 1). Importantly, during reaching the range of EMG changes is larger than across the different postures, yet we did not observe reduced slopes of the regression during reaching movements, which would result from the saturation of MEPs at higher EMGs. However, the MEP gains for distal muscles [Br, FCR, FCU, and extensor carpi radialis (ECR)] were rather high, which suggests a very high recruitment of these projections by TMS. Therefore, the potential saturation of corticospinal contribution to the recruitment of distal muscles probed with TMS in our study cannot be completely ruled out. Fortunately, this would not change our results and conclusions that are based primarily on strong relationships between the corticospinal contributions and the activity of eight proximal muscles and postural torques about the shoulder and elbow.

Finally, we cannot exclude the possible contribution of fatigue to changes in MEP characteristics. However, both the movements and stimulation timing were pseudorandomized specifically to prevent systemic biases, such as fatigue, from influencing our results. Furthermore, session 1 was always done last because it was the easiest. Despite this systematic bias, there were no systematic changes between static and dynamic MEPs in the same muscle (Fig. 7). In some muscles MEPs were larger in session 1 compared with session 2, while in other muscles they were smaller. If fatigue was systematically affecting MEP magnitudes, we would expect a more generalized trend such as the slope between the static MEPs and background EMG always being larger or smaller than the slope between dynamic MEPs and background EMG in at least one movement direction across all muscles and individuals. Fatigue would also most likely affect the muscles that are contracting the most, i.e., antigravity muscles like the anterior deltoid and biceps. However, the MEPs in anterior deltoid were larger in Static compared with Control conditions, whereas the MEPs in biceps were smaller or the same in Static compared with Control conditions (Fig. 7). Altogether, these results and our experimental design assure us, and we hope the reader, that fatigue was an unlikely contributor to our study’s primary findings.

DATA AVAILABILITY

All data relevant for assessing the conclusions of this study are presented in the text and figures. The code to calculate joint angles from motion capture and the dynamic model of the arm used for inverse simulation is uploaded onto https://github.com/NeuroRehabLab/Limb-Dynamics. Less processed deidentified data are available upon request.

SUPPLEMENTAL DATA

Supplemental Figs. S1–S5: https://doi.org/10.6084/m9.figshare.21383796.v1.

Supplemental Figs. S6–S10: https://doi.org/10.6084/m9.figshare.21383781.v1.

GRANTS

R.L.H. was supported by training grants T32GM081741 and T32AG052375 from the National Institute of General Medical Sciences (https://www.nigms.nih.gov), research grants NIH Grant P41 EB018783, NYS SCIRB C32236GG and C33279GG, and the Stratton VA Medical Center. V.G. was supported by Grants P20GM109098 and P30GM103503 from the National Institute of General Medical Sciences and 1007978 from the National Science Foundation.

DISCLAIMERS

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

P.H.E. and V.G. conceived and designed research; R.L.H. and V.G. performed experiments; R.L.H. and V.G. analyzed data; R.L.H., P.H.E., and V.G. interpreted results of experiments; R.L.H. and V.G. prepared figures; R.L.H., P.H.E., and V.G. drafted manuscript; R.L.H., P.H.E., and V.G. edited and revised manuscript; V.G. approved final version of manuscript.

ENDNOTE

At the request of the authors, readers are herein alerted to the fact that additional materials related to this manuscript may be found at https://github.com/NeuroRehabLab/Limb-Dynamics. These materials are not a part of this manuscript and have not undergone peer review by the American Physiological Society (APS). APS and the journal editors take no responsibility for these materials, for the website address, or for any links to or from it.