Journal Logo


Strength Training Biases Goal-Directed Aiming


Author Information
Medicine & Science in Sports & Exercise: September 2016 - Volume 48 - Issue 9 - p 1835-1846
doi: 10.1249/MSS.0000000000000956
  • Free


Strength training can alter the neural circuitry involved in generating the training movements, resulting in rapid strength gains before substantial muscular adaptation (7). Evidence for such changes can be observed after a single session of isometric strength training, via changes in the direction of muscle twitched evoked by transcranial magnetic stimulation (TMS) (30). This implies that strength training can alter neural connectivity within the human primary motor cortex or its downstream projections (30,36). These findings are similar to those observed in response to extensive practice of simple ballistic finger tasks (4,8), suggesting that responses to strength training may be brought about by a form of “use-dependent learning.” If this is the case, then strength training should also cause behavioral effects that have been attributed to use-dependent learning. For example, the repetition of a movement can bias subsequent movements to resemble the characteristics of the repeated action. Both the direction and the speed of movements are affected by movement history (16,37), and Verstynen and Sabes (37) showed through neural network simulation how such biases could emerge from “trial by trial” learning processes based on Hebbian principles. Recent data also illustrate that use-dependent learning contributes to the final solution of error-based sensorimotor adaptation tasks through processes that are distinct from the well-established processes of motor adaptation (10,18). It remains to be tested whether early neural adaptations to strength training also bias subsequent voluntary movements toward the direction of movements executed during training, although strength training can alter muscle coordination in more complicated sensorimotor tasks (5).

The degree to which bias effects generalize to different contexts can reveal how neural adaptation to strength training is represented. Generalization characteristics have been documented for error-based forms of motor learning in the context of sensorimotor adaptation paradigms. For example, adaptation to visuomotor rotation generalizes narrowly and only affects movements directed within ±45° of the training movement direction. Similarly, the adaptation of limb dynamics also generalizes narrowly, as shown by exposure to a novel force field or altered inertial properties of the arm (10,38). Generalization patterns can also provide inference about the form in which learning is represented in the brain. Classic studies suggested that visuomotor rotation and dynamic learning are represented in extrinsic (23) and intrinsic (31) coordinates, respectively. However, more recent work suggests that both forms of learning involve a more integrated representation than originally thought, potentially involving multiple coordinate frames (2,3). Generalization characteristics have not been systematically documented for use-dependent learning or strength training. Thus, we hope to capture this information at the earliest stages of strength training and provide a reference for longer-term strength training studies.

We sought to identify the characteristics of any bias in force aiming produced by strength training in two separate experiments. In the first experiment, we investigated if a low-force isometric forearm aiming task produced a bias in force direction in various areas of the workspace (16 targets) in response to a maximal-force ballistic contraction in a single training direction. In the second experiment, we determined if a strong neural drive was necessary to induce such a bias on eight targets by comparing the outcome of the maximal-force ballistic contraction in condition 1 to a low-force contraction in condition 2. Both the aiming task and the training were conducted in the same (neutral) forearm posture for conditions 1 and 2. In addition, we tested whether this bias is represented in an extrinsic or muscle-based reference frame in condition 3 by performing the maximal-force ballistic contraction in a 90° pronated forearm posture, whereas the low-force aiming task was performed in a neutral forearm posture. This design allowed us to test if the bias shifts toward training direction in extrinsic space or toward the trained muscles.



All participants (10 [6 males and 4 females with a mean age of 27.1 ± 6.7 yr] for experiment 1 and 12 [7 males and 5 females with the mean age of 25.6 ± 4.4 yr] for experiment 2) recruited for these experiments were without a history of neurological disease or previous strength training experience specifically for the wrist muscles and were right-handed (confirmed by the Edinburgh handedness questionnaire). All participants provided their written informed consent to the procedures, which conformed to the Declaration of Helsinki and was approved by the University of Queensland medical research ethics committee. None of the participants took part in both the studies. The experiments were designed to examine if a bias occurred in the force direction of an isometric forearm aiming task in various areas of the workspace after a maximal-force ballistic contraction in a single training direction (experiment 1) and to determined whether this bias is represented in an extrinsic or muscle-based reference frame and if a strong neural drive influences the degree of bias (experiment 2).

Setup and force measurement

Participants were seated on a comfortable chair with the forearm secured in a custom made wrist clamping device. The forearm was held either in a neutral (i.e., midpoint between supination and pronation; see Fig. 1A) or pronated posture during force measurement. The hand was secured with 12 adjustable surface clamps at the metacarpophalangeal joint, and the forearm was secured with 10 clamps just proximal to the wrist to prevent movement in the device. Polyurethane foam straps (thickness = 0.3 cm) were used to pad the contact locations. The device could be rotated about the long axis of the forearm (i.e., in pronation–supination) without a requirement to adjust the surface clamps. A six-degree-of-freedom force transducer (JR3 45E15A-I63-A400N60S, Woodland, CA) registered radial–ulnar deviation and flexion–extension forces at the wrist. A computer screen was placed 1.5 m from the participants at eye level to provide online visual feedback of the task. A cursor corresponded to the wrist force vector in a two-dimensional display oriented in the frontal plane. Force data were sampled at 2 kHz with a 16-bit National Instruments A/D board operated by a computer running a custom-written Labview program (LabVIEW, Version 8.2.1; National Instrument, USA). All data were analyzed offline using a custom-written Labview program.

A, Schematic illustration of experimental setup including visual display location and the neutral forearm posture in the wrist clamping device. B, Distribution of target locations for the isometric aiming task in experiment 1. The repeated force direction (315°) is specified with a black vector. C, Distribution of target locations for the isometric aiming task in experiment 2. For the pronated training condition, the repeated force direction (35°) in extrinsic space when the wrist was trained in the pronated posture is specified with an open circle, and a closed black circle identifies the corresponding repeated force direction in muscle- and joint-based coordinates when the wrist was tested in the neutral posture (305°). For the high and low-force conditions, in which training and tested were performed in a neutral posture, the closed black circle specifies the repeated direction in all relevant reference frames. D, Schematic representation of the study design in experiment 1, which began with MVC in the training direction (not shown) followed by two familiarization sessions. Pretest consisted of five sets of probe trials to each of the 16 targets, whereas posttest at 30 s and 10 min consisted of one set and five sets, respectively. Strength training involved high-force contractions to a 315° target. Both testing and training were conducted in the neutral forearm position. E, Schematic representation of the study design in experiment 2. The testing schedule was similar to experiment 1 and performed in the neutral forearm position, with three sets of eight probe trials at pretest and after 10 min, and one set at after 30 s. Training consisted of three conditions performed at least 1 wk apart; condition 1 (C1) and condition 2 (C2) involved high and low-force training respectively, performed in the neutral forearm position to a 305° target. Condition 3 (C3) involved high-force contractions performed in the pronated forearm posture to a 35° target.

Aiming task

The isometric force aiming task required participants to move the cursor from the center of the two-dimensional display to radial targets situated 75% of the distance from the origin to the edge of the display (10 cm) within a movement time of 150–250 ms (defined as the time to move the cursor from 10% to 90% of target distance). Cursor gain was held constant for each person such that 30 N (n = 15) or 25 N (n = 7) was required to move from the origin to the edge of the display. A high-pitched tone (500 ms, 800-Hz sinusoid) signaled successful target acquisition when the cursor was continuously within the target radius (10% of distance from origin to center of target) for 50 ms, at which time the cursor and the target were simultaneously extinguished. A second tone was sounded 500 ms later to indicate whether movement time was correct (high-pitched tone) or not (low-pitched tone; 500 ms, 300-Hz sinusoid). A color-coded (green = correct, red = too fast or slow) vertical bar indicated the movement time on the last trial. In familiarization trials, the cursor was visible for a random fore period (1–2 s) before target presentation until target acquisition, or until 1.7 s after target presentation (if the target was not acquired). In probe trials, an expanding circle concentric with the origin indicated the force magnitude without directional information and was visible for equivalent periods. Target acquisition was defined as when the radius of the expanding circle intersected the target radius irrespective of the force direction, and identical auditory and visual feedback was provided. A full cycle of one trial to each target was presented before any target repetition, and targets appeared in random order within each cycle.

Strength training tasks

At the beginning of each session, each person’s maximal voluntary contraction (MVC) force was measured as the peak force attained in three trials toward a target located at the training direction to calibrate force targets for the training tasks. Cursor gain was set to 300 N from origin to display edge. The strength training task involved four blocks of 10 contractions to a single target, with 2 s between each contraction and 3 min between blocks. A reference white vector from the origin to the target force level provided the participants with the specified training direction, and a red vector from the origin to the current force provided online feedback of force direction and extent. For ballistic training, the reference white line was set to 100% MVC, with cursor gain of 300 N, and participants were required to contract as fast and hard as possible and sustain a 2-s maximum effort isometric voluntary contraction with minimum possible deviation between force feedback and reference vectors. A brief tone was used to signal the requirement for rapid force production (400-Hz sinusoid, duration = 0.25 s), which was followed by a continuous tone of 2 s duration (900-Hz sinusoid). For the low-force training task, participants were required to perform a slow-ramp contraction to reach 10% MVC for a period of 1 s (signaled by a sinusoidal tone with linearly increasing frequency from 400 to 535 Hz) and then to sustain the contraction for a further 2 s before relaxing (signaled by a continuous tone of 2 s duration, frequency = 1.2 kHz). Cursor gain was set to 30 N.

Experiment 1

In each aiming task block, participants were asked to move the cursor to 16 targets distributed symmetrically about the origin (i.e., 22.5° apart; see Fig. 1B) with the forearm in the neutral position. Figure 1D provides a schematic representation of the task schedule. After MVC assessment, participants first completed an aiming task block involving 96 familiarization trials with full cursor feedback of force direction (six trials per target). They then performed another block of 160 trials in which familiarization and probe trials (force magnitude feedback only) were alternated for each full cycle of targets (five trials per target for each trial type). Baseline aiming directions were then established in a block of 80 probe trials (five trials per target). All participants completed the ballistic training task with a reference direction of 315°. After training, two blocks involving only probe trials were performed. The first block of 16 trials was initiated 30 s after the completion of training (one trial per target). The second block of 80 trials began 10 min after training (five trials per target).

Experiment 2

All participants completed three testing conditions in a counterbalanced order, each on different days separated by a minimum of 1 wk. In each condition, they performed force aiming contractions to eight targets, oriented to probe bias surrounding two potential training directions (35° and 305°, see Fig. 1C). Targets were located both to align with and oppose the training directions by 180° (i.e., 35°, 125°, 215°, and 305°) and surround the training directions with a 15° offset on either side (20°, 50°, 290°, and 320°). The training directions were chosen to align approximately with the preferred activation directions for the extensor carpi ulnaris muscle with the forearm maintained in a neutral position (305°) and pronated position (35°) (9). The schedule of testing was similar to experiment one (see Fig. 1E), except that the first aiming block involved 40 familiarization trials (five trials per target), the second aiming block involved 56 familiarization and probe trials (alternating by cycles of eight trials, three cycles of probe trials, and four cycles of familiarization trials), the baseline assessment involved 24 probe trials (three trials per target), and the performance posttraining was assessed at 30 s with eight probe trials (one trial per target) and at 10 min with 24 probe trials (three per target).

The aiming task was performed in the neutral forearm position in all three experimental conditions. In two of the three conditions, both the aiming task and the training were conducted with the forearm in the same (neutral) posture (see Fig. 1A). One of these conditions involved weak forces (slow ramp to 10% MVC) and the other involved maximal-force ballistic contractions. The training direction for both sessions was 305°. However, in the third condition, maximal-force ballistic contractions were performed toward a 35° training target with the forearm pronated 90° relative to neutral. This meant that the probe target directions and any training-induced bias could be defined according to the training direction in extrinsic space (35°) or joint space (305°). The suitability of the wrist joint system for this purpose has been recognized previously, as simple forearm rotation allows dissociation between extrinsic and muscle and/or joint coordinate frames (e.g., 9,20). Critically, we have shown that preferred directions of wrist muscle activation and muscle pulling directions follow closely the orientation of the wrist joint with forearm rotation when posture is carefully controlled (9). Thus, in contrast to previous reports in nonhuman primates (20,38), joint and muscle-based coordinates are similarly affected by forearm rotation in humans. Consequently, our current experiments could not be confounded by putative mismatches between training target direction in muscle and joint space. Moreover, the isometric task we used requires a relatively simple visuomotor transformation between the difference vector between cursor and target location and the required wrist force vector. This excludes potential confounds associated with changes in limb dynamics and spatial location relative to the body that can accompany postural manipulations designed to dissociate spatial frames of reference.

Basis function model

A simple, deterministic basis function model was used to simulate the transformation of a target direction θT into a force vector via a population of N directionally tuned neurons.

Each neuron i is defined by a preferred direction θPD,i at which its firing rate ri is maximal and a weight vector

, which jointly specifies its contribution to the output force vector. The firing rate of neuron i is defined either by a truncated cosine,

where [x]+ returns x for x > 0 and 0 for x ≤ 0 or by a Gaussian function of width σ,

In this later case, the Gaussian function was made periodic for the 360° range. The output force vector is then defined by vectorial summation:

In the absence of adaptation, each neuron’s weight vector is a unit vector that points toward its preferred direction, such that the resulting direction of the output force matches the direction of the input target:

We modeled the adaptation induced by movement repetition to a single target direction by an increase in the length of weight vectors of each neuron that is proportional to the firing rate of that neuron in response to the training movement direction:

where α is a scaling factor chosen to enforce the peak bias predicted by the model to equal the maximal experimentally measured bias in each condition. Simulations were conducted with N = 360 neurons, θPD,i evenly distributed (one neuron per degree). We varied the free parameter σ to best fit experimental data. The value of α is uniquely defined for each value of σ, and in that sense, it is not a free parameter. Basis function models of this type have been successfully used to predict the direction of movement from population recordings of single neuron activity in M1 and the generalization of motor learning (13).

Data analysis

The force direction for each aiming task trial was determined as the angle between the rightward, horizontal target (i.e., at 3 o’clock) and a vector joining the origin and the cursor at 95% of the target distance. All trials in which the force direction differed from the target direction by greater than 45° were discarded as outliers (1.3% of trials in experiment 1, 0.7% of trials in experiment 2). The direction of each training trial was determined as the mean cursor direction during the maximal-force phase of the contraction or the steady 10% contraction in the low-force training condition. For all subsequent analyses, target directions were expressed as the absolute angle relative to each person’s mean training direction. The absolute directional errors between force directions and target directions were recorded at baseline, 30 s, and 10 min posttraining. Posttraining directional errors were defined such that biases from baseline toward the training direction were considered positive, and biases from baseline away from the training direction were considered negative. To determine whether training caused systematic errors, plots of mean biases versus absolute angle from the training direction were fitted with a quadratic linear regression model (model: y = ax2 + bx + c). The SD values of force direction for each subject and target were computed for the five baseline aiming blocks, and the five posttraining aiming blocks initiated 10 min after training. These were analyzed with a two-way repeated-measures ANOVA (target × time: baseline vs 10 min posttraining).

To determine whether there was structure in the bias data approximating that predicted by altered output weightings of the basis function model, we calculated the mean sum-squared errors in bias between the model prediction and the experimental data and compared this to the mean sum-squared errors in bias between 10,000 randomly drawn samples of the eight possible permutations of target direction and observed bias. The basis functions of the individual neural elements in the model were set to the Gaussian width that provided the best fit to the experimental data, a narrowly tuned Gaussian (σ = 23°) (35), or a truncated cosine. The model fits were considered significant if the mean sum-squared error of the fits to the experimental data was within the lowest 5% of mean sum-squared errors for fits to the permuted data sets. The proportion of the variance in aiming bias explained by the model fits relative to the mean bias (r2 = 1 − SSEfit / SSEmean) was also calculated. Note that model fits poorer than the mean bias yield negative r2 values.

For each condition in experiment 2, plots of mean biases versus absolute angle from the training direction were also fitted with the quadratic linear regression model. In addition, mean biases from baseline force direction were calculated for six of the eight targets, excluding the target that aligned with the training direction and the target 180° opposite. Two mean bias values were calculated for the condition in which training was performed with the forearm pronated, such that the training direction was defined either in joint space (i.e., 305° on the visual display) or in extrinsic space (35° on the visual display). A two-way repeated-measures ANOVA (four training conditions × two measurement times) was used to detect if the mean bias differed between conditions. A single missing cell due to an excluded outlier was replaced by the group mean bias for the low-force condition. To determine whether ballistic training induced bias according to an extrinsic or joint-based reference frame, a pairwise planned contrast was calculated between the extrinsic and the joint-based errors. To determine whether the degree of bias induced by repeated force production is affected by strong neural drive, a pairwise planned contrast was calculated between the ballistic training and the weak force errors. The Kolmogorov–Smirnov test (using Lilliefors probabilities for sample-based estimates of mean and SD) showed that the distributions of all variables were approximately normal. The Mauchley test was used to assess the sphericity of repeated-measures effects with greater than two levels, and the Greenhouse–Geisser degree of freedom corrections was applied to cases where the assumption of sphericity was violated. The alpha was set at 0.05 for all comparisons.


In experiment 1, we measured isometric aiming errors induced by 40 maximal-force contractions initiated as fast as possible to a single radial target. Figure 2 shows the directional errors between force directions and target directions at baseline, 30 s, and 10 min posttraining. The baseline plot shows an inherent bias that varies with different target directions. This bias resembles the well-described “oblique effect” that has been reported previously in studies concerning the perception of line orientation and execution of target-directed arm movements (32,33). Our results are consistent with a summation of this inherent bias with an additional effect because of maximum-force ballistic contractions. Note that errors at posttraining both amplify and reduce the baseline bias depending on the target location with respect to the training direction. Thus, a simple magnification of the inherent bias cannot explain our results.

Plot of group mean ± SEM absolute directional errors between force directions and target directions for baseline (solid black), after 1–30 s after training (dashed black), and after 2–10 min after training (dashed gray) for each of the 16 targets specified as absolute angle from the repeated force direction.

In Figure 3, we express the bias after training relative to the baseline aiming direction, with errors toward the training direction positive and errors away negative. Our results reveal that forces applied soon (i.e., 30–110 s) after the repeated high-force contractions were biased toward the training direction when subjects aimed at targets distributed broadly throughout the radial workspace (Fig. 3A). This observation is supported by a significant quadratic linear regression fit (F3,5 = 15.6, r2 = 0.86, P = 0.0007). The bias effect was substantially reduced by the first block of 16 aiming trials initiated 10 min after training (Fig. 3B) such that the quadratic regression was nonsignificant (F3,5 = 1.71, r2 = 0.41, P = 0.27). Results were similar if the mean bias over all five probe trial blocks initiated at 10 min posttraining was considered rather than the bias from only the first block (F3,5 = 0.99, r2 = 0.28, P = 0.44). There was no significant difference in the variability of force directions across the five trials to each target direction between the baseline and the 10-min posttraining probe trials (F1,9 = 0.006, P = 0.94), nor a significant probe time by target direction interaction effect (F15,135 = 0.94, P = 0.52). Thus, our experimental design did not provide evidence for the variance–bias tradeoff previously reported to accompany movement repetition (16,37).

A, Group mean ± SEM bias from baseline force direction for each target specified as absolute angle from the repeated force direction at 30 s after training. C, Model fits to the bias distribution are shown for neural basis functions constrained to be narrow Gaussians (23° solid line) and truncated Cosines (dashed line), or with the Gaussian width adjusted to best fit the data (44° dotted line). *Significant fits (see text). B, Group mean ± SEM bias from baseline force aiming direction for each target specified as absolute angle from the repeated force direction for the first trial at 10 min after training. D, Model fits were not significant.

The unimodal distribution of bias induced to targets presented throughout the workspace, revealed by significant quadratic linear regression analysis, was also captured by the basis function network model. The predicted bias best matched that observed experimentally when the model was composed of relatively wide Gaussian basis functions (σ = 44°, r2 = 0.80, P = 0.0015). Truncated cosine basis functions also provided a significant fit to the data (r2 = 0.80, P = 0.0012). By contrast, narrow Gaussian tuning widths shown previously to best represent the generalization of learned visuomotor rotation (35), provided a poorer fit to the experimentally observed bias induced by movement repetition (σ = 23°, r2 = −0.33, P = 0.067) (Fig. 3C). The model fits were not significant in the first block of 16 aiming trials initiated 10 min after training (Fig. 3D).

In experiment 2, we investigated whether aiming bias induced by repeated contractions to a single direction generalizes according to an intrinsic or extrinsic coordinate frame and whether high-force exacerbates bias. Group mean bias values averaged across all targets for the different conditions are presented in Figure 4. Participants performed three training conditions on different days in which they performed 1) high-force and 2) low-force contractions with the forearm in a neutral position and 3) high-force contractions with the forearm pronated. For the pronated training condition, bias was probed with the forearm neutral so that it could be measured with respect to the repeated force direction either in extrinsic (i.e., 35°) or joint-based (i.e., 305°) coordinates. In the high-force condition, forces applied by subjects aiming at eight targets biased toward the training direction 30 s after training and produced a significant fit (F3,5 = 5.79, r2 = 0.70, P = 0.05) as illustrated in Figure 5A. This bias appeared to be partially retained after 10 min of training (Fig. 5B) but was not strong enough to produce a significant quadratic regression fit (F3,5 = 4.49, r2 = 0.64, P = 0.077). Although there was a general reduction in the magnitude of bias after 10 min, the apparent partial retention of bias when aiming was tested 10 min after the completion of training in this second experiment contrasts with the results of experiment 1. Given that experiment 1 involved double the number of probe actions at 30 s posttraining, this discrepancy indicates that the execution of actions to probe targets contributed to the washout of bias. This occurred although subjects received no feedback of their aiming performance during probe trials. Thus, it would appear that the effects early after strength training dissipate with additional action in addition to any influence of the passage of time.

Group mean ± SEM bias from baseline force aiming direction in experiment 2, pooled across targets, at 30 s and 10 min after training. Pooled across measurement time points, bias was significantly greater for the high-force than low-force training, and for the extrinsic than the muscle-based target specifications.
Group mean ± SEM bias from baseline force direction for each target specified as absolute angle from the repeated force direction in experiment 2 at 30 s after high-force training (A), 10 min after high-force training (B), 30 s after low-force training (C), 30 s after high-force training expressed in extrinsic space coordinates (D), and 30 s after high-force training expressed in muscle space coordinates (E).

The mean bias for six targets pooled across both probe times was substantially greater after subjects performed 40 ballistic maximal-force contractions than after they performed the same number of weak contractions with a slow-ramp force onset (F1,11 = 5.79, P = 0.03). Slow-ramp force contractions also did not produce a significant quadratic regression fit (F3,5 = 0.63, r2 = 0.20, P = 0.57) (Fig. 5C). This suggests that the high-force or ballistic nature of the training actions exacerbates biases induced by movement repetition.

For the condition in which posture was altered to assess the coordinate frame of generalization, bias was significantly greater when expressed relative to the extrinsic than the intrinsic training direction (F1,11 = 7.71, P = 0.02). The quadratic linear regression also produced a significant fit for the bias expressed in extrinsic space coordinates (F3,5 = 11.2, r2 = 0.82, P = 0.014) (Fig. 5D) but not for the muscles space coordinates (F3,5 = 0.47, r2 = 0.16, P = 0.65) (Fig. 5E). This indicates that the direction of target-directed forces tended to be drawn toward the direction of previously executed actions in extrinsic space, rather than to the pulling direction of the trained muscles. This clearly rules out a peripheral mechanism such as muscle fatigue or potentiation for the bias effects in the current study, and informs speculation as to the possible underlying neural substrate (see Discussion). Critically, for both of the two targets located within the minor arc between the extrinsic and intrinsic training directions (i.e., 20° and 320°), the mean bias at 30 s posttraining was toward the repeated force direction in extrinsic coordinates (mean and 95% CI = 10.6° ± 5.6°), which reflects the observation of extrinsically oriented aiming errors on 20/24 of the individual trials to these targets across all participants.


The data illustrate the effects of early neural adaptation to strength training on the direction of isometric forces aimed toward visual targets. This was conducted by measuring biases in force directions to a broad distribution of targets after a single session of maximum-force ballistic training. In experiment 1, we found that bias toward the training direction occurred for targets distributed broadly throughout the workspace immediately after training but was absent 10 min after training. In experiment 2, we showed that high-force ballistic contraction was necessary to induce such a bias because low-force contractions did not reveal a systematic bias toward the training direction. The bias that was induced by high-force contraction immediately after training was also partly retained 10 min after training. In addition, we showed that the bias toward the training direction was expressed according to an extrinsic rather than muscle-based reference frame. These findings provide insight into the computational modules and likely neural substrates that underlie early neural adaptation to strength training.

The distribution of bias observed in experiment 1 immediately after training approximated a skewed unimodal distribution, with a peak in the range 50°–80°, and a return to zero bias at approximately 135° (Fig. 2A). The broad and skewed unimodal distribution of bias experimentally observed here was well approximated by the simplest of learning rules (i.e., a linear scaling of output weights according to activation during repeated movement) applied to the simplest of neural network models (i.e., a single layer of neurons with firing rates tuned to target direction and output force direction determined by a weighted sum of the preferred directions of firing neurons). This illustrates that the reinforcement of synaptic connections between directionally sensitive neurons active during repeated movement and their downstream targets can produce qualitatively similar bias distributions to those experimentally observed. Model fits to the data were best when neurons were tuned according to broad basis functions, which calls to mind cosine tuning of neurons in primary motor cortex (M1, e.g., [13]). Although other frontal and parietal sensorimotor areas contain neurons that appear broadly tuned to movement direction, we also observed greater aiming bias after repeated action involving high forces than after weak repeated actions. This indicates that areas involved in the specification of muscular forces are involved in our behavioral effects and suggests a contribution from high neuronal firing rates associated with strong force or rapid acceleration. M1 firing rates are strongly related to contraction force (11,12), whereas the firing of neurons in parietal area 5, which appears to be involved in motor planning and contains neurons that display broad spatial tuning, correlates better with the kinematics than the force requirements of movement (15).

The results at 10 min posttraining, however, did not reveal significant bias from pretraining force directions. This outcome could be due to a decay in the effect with the passage of time or to the execution of multiple isometric contractions throughout the workspace during this period (37). In a separate study, a brief time course was also observed for errors in estimating force magnitude after repeated contraction. The outcome was also attributed to the decay in the effect over time (19). However, the results of experiment 2, in which we reduced the number of targets to eight, suggest a role for additional factors beyond a mere decay in the effect over time. The fact that fewer movements were executed to probe bias after training in experiment 2 (i.e., 8 at 30 s and 24 at 10 min in experiment 2, vs 16 at 30 s and 80 at 10 min in experiment 1) suggests that bias effects dissipate with the execution of additional actions, even in the absence of overt visual feedback about directional errors.

A key question addressed in experiment 2 was whether use-dependent aiming biases are tied to the direction of repeated action in extrinsic space, or to the requisite muscle activations and joint forces. The data indicate that the average bias toward the extrinsic direction of repeated actions was similar when the same posture was maintained in training and testing, compared with when the forearm was rotated to different postures for training and testing. By contrast, mean bias was negligible toward the pulling direction of the muscles involved in the repeated actions when this was dissociated from the repeated (extrinsic) target direction by forearm rotation. This suggests that the bias effect is mediated in a computational module concerned with the movement plan in extrinsic space. By contrast, recent work suggests that both visuomotor and dynamic adaptation likely involve integration across multiple frames of reference (2,3). Moreover, our previous study (5) found that 4 wk of training conferred a performance benefit in muscle-based coordinates on a difficult timing task requiring the coordination of agonist and antagonist muscles. Note, however, that there was no spatial accuracy component to this task, and that the possibility that training also produces extrinsically referenced performance benefits was not tested. An apparently exclusive extrinsic generalization of aiming bias in the current study is therefore surprising. It is possible that this observation was due to the tight control over the manipulation of spatial reference frames in the current study, such that the potential involvement of factors such as changes in limb dynamics and the spatial location relative to the body were excluded. However, there is also a potential confound in the experimental design of experiment 2, whereby four of the six targets used to measure bias were situated clockwise to the training extrinsic direction. Thus, any global change in actual or felt wrist position, due to changes in wrist posture from neutral to pronation during the experiment, would asymmetrically affect the bias averaged for the six targets. Thus, it is possible that such an effect exacerbated the apparent extrinsic bias observed. However, subtracting the mean errors from baseline across all eight targets before calculating bias did not change the pattern of results and only reduced the size of the effect.

Finally, although pure extrinsic generalization can be reconciled with a putative M1 locus for the circuits underlying strength training, the firing of M1 neurons can correlate either with the extrinsic direction of motion, or follow the pulling direction of muscles upon forearm rotation (20). In this regard, synapses onto M1 neurons from axons originating in secondary motor areas such as ventral premoter cortex, where neuronal firing appears broadly tuned to the direction of motion defined exclusively according to extrinsic space (21), appear a likely candidate for the site of adaptation. However, our data would appear to argue against a substantial contribution of spinal circuits to the current bias effects (see Giesebrecht et al. [14]), given that spinal interneuron firing appears to be strongly coupled to the direction of muscular action (38). By contrast, our data from experiment 2 show bias away from the pulling direction of the exercised muscles. In sum, our data suggest that strength training affects computational modules concerned with the motor plan in extrinsic coordinates and may involve mechanisms that resemble synaptic potentiation acting on connections onto M1.

The findings in experiment 2 also provide new evidence that strong contractions exacerbate use-dependent bias effects. The simplest possible explanation for this is that the high neuronal firing rates in motor cortex associated with the generation of high forces (11,12) exacerbate synaptic potentiation (24). However, it is also possible that the greater effort required to produce maximal ballistic contractions enhanced the reinforcement of the repeated action through reward or arousal related processes. Although we cannot rule out a reward-based contribution to the bias effects, the particulars of our current experimental task argue against this interpretation. Specifically, in both conditions, subjects were instructed to match the target force vector as closely as possible, but there was no explicit feedback of task performance. The number of putative action–reward associations was identical between the maximal- and low-force training conditions. Moreover, although less muscular effort was required to reach the target in the low-force condition, the requirement to accurately maintain force at a specific end point in two-dimensional force-space necessitated constant attention. Thus, we favor a direct contribution from neuronal firing rate to the strength training effects observed.

An important additional question arises from the comparison between the maximum-force ballistic contractions and the slow-ramp low-force contractions. Was it the maximum force, high drive component of strength training, or the ballistic, high rate of force development component that induced bias? Both components have been associated to the high neuronal firing rates; however, a recent study found that training involving either ballistic or slow-ramp contractions produced similar facilitation of motoneuron responsiveness (27). In addition, we previously showed that changes in the magnitude and direction of TMS-evoked twitches were comparable between ballistic and slow-ramp contractions (30). Therefore, it would appear that the production of high neural drive is a critical component required to induce corticospinal responsiveness effects and may also underlie the behavioral bias effects found in the current study.

An important issue to consider is whether the transient aiming biases we observed are of functional relevance in practical contexts. In this regard, it is important to note that the primary training outcome of strength training, namely, increased strength, is not immediately apparent after a single training session. Rather, the adaptive processes that ultimately increase strength remain latent until the training stimulus has been repeated sufficiently to accumulate into functionally observable effects. Thus, although the functional benefit of training is not immediately apparent after a single session, each session contributes to the longer-term strength gains. This is illustrated by the results of study in which low-frequency repetitive TMS applied immediately after every session of a 4-wk training program resulted in impaired strength gains (17). Moreover, our previous work shows that there were changes in TMS-evoked twitch direction after a single session of training (30), and parallel twitch direction effects were observed after 4 wk of training (25). Rosenkranz et al. (29) also showed that ballistic skill learning, which is known to induce changes in TMS-evoked twitch direction (9), results in continued improvement in performance with repeated training bouts. In sum, it seems clear that subtle and transient changes observed after single strength and ballistic training sessions have important functional consequences. We therefore believe that our findings are functionally relevant, especially to integrated training involving various strength and skill tasks within a session.

Mechanism of adaptation

The speculation that M1 is a crucial substrate for strength training is supported by previous work using the same strength training intervention, in which the direction of muscle twitches evoked by noninvasive stimulation of motor cortex was altered (30). The findings of this study were similar to those reported after the use-dependent learning of a ballistic finger task (8). Use-dependent learning has been proposed to occur predominantly at the motor cortex, on the basis of studies using brain stimulation (8,26,28), pharmacological methods (40), and neural network simulation (37). Moreover, Nuzzo et al. (27) recently reported increases in muscle responses evoked both by TMS and stimulation of the corticospinal tract at the cervicomedullary junction. The authors highlighted that increased responses to brain stimulation were likely to be dominated by spinal level factors, although precise contributions were difficult to assess because a direct comparison was prevented by differences in response amplitude. More interestingly, however, the authors found that the time course of changes in motor-evoked potentials and cervicomedullary motor-evoked potentials was different after training. The increase in the responses to brain stimulation were largest immediately after training and remained facilitated throughout the experiment (30 min), with a gradual trend toward pretraining levels. By contrast, responses to corticospinal tract stimulation only increased after 5 min of training and remained elevated for a further 15–20 min, suggesting that at least some of the increase in cortical responsiveness after strength training can be attributed to cortical changes. This finding shows that cortical changes are likely to occur immediately after strength training, a time course similar to our study, and thus support the case for changes occurring in M1.

In our previous study, we proposed that strengthening of connections within neural networks in favor of the training direction occurs early in strength training (30,36). Here we extend knowledge of the acute responses to strength training to show that such effects can affect voluntary aiming behavior. The acute neural changes underlying both behavioral and responsiveness effects may be due to changes in synaptic efficacy or excitability of neurons as a result of long-term potentiation (LTP) and may precede additional long-term structural changes in neural connectivity that underlie relatively permanent learning (22,29). LTP appears to be a likely mechanism involved in this early neural adaptation to strength training, based on evidence from studies that recorded a shift in TMS-evoked twitches toward the direction of repeated ballistic movement (4,8). For example, Bütefisch et al. provided pharmacological evidence by manipulating the N-methyl-D-aspartate and γ-aminobutyric acid A receptors (for a detailed description, see Butefisch et al. [4]). Converging evidence that LTP is implicated in use-dependent learning was provided by Ziemann et al. (39) and Rosenkranz et al., (29) in studies that used repetitive noninvasive brain stimulation techniques to induce cortical plasticity (i.e., paired-associative stimulation [34]).


The capacity of strength training to influence the characteristics of related movements that were not specifically trained is of primary practical importance because most strength training programs aim to improve performance on functional tasks beyond the direct training environment. We have previously conceptualized such generalization as a “transfer of learning”(6). Our current data show that the effects of strength training can transfer to related behaviors after a single session of training, whereas most previous studies have focused on the effects of a few weeks of training (i.e., 1,5). We also showed that the effects of training generalize broadly, such that bias toward the training movement direction can occur even for movements orthogonal to the training direction. This is even the case for movements in which the primary agonist muscles were not targeted by training. The observation that adaptation to strength training appears to drive subsequent movements toward the training direction, regardless of the muscles that are involved, might have practical implications for training design. Although the functional importance and persistence of the effect require further investigation, the data suggest that caution should be applied when multiple strength and/or skill tasks are completed within a single training session.

The authors thank David Lloyd for the figure of the apparatus. This research was supported by the High Impact Research Chancellory (grant nos. UM.C/625/1/HIR/214 and RG529-13HTM) from the University of Malaya and the Australian Research Council (grant nos. DP1093193 and FT1201003890). The results of the present study do not constitute endorsement by the American College of Sports Medicine.


1. Barry BK, Carson RG. Transfer of resistance training to enhance rapid coordinated force production by older adults. Exp Brain Res. 2004;159(2):225–38.
2. Berniker M, Franklin DW, Flanagan JR, Wolpert DM, Kording K. Motor learning of novel dynamics is not represented in a single global coordinate system: evaluation of mixed coordinate representations and local learning. J Neurophysiol. 2014;111(6):1165–82.
3. Brayanov JB, Press DZ, Smith MA. Motor memory is encoded as a gain-field combination of intrinsic and extrinsic action representations. J Neurosci. 2012;32(43):14951–65.
4. Butefisch CM, Davis BC, Wise SP, et al. Mechanisms of use-dependent plasticity in the human motor cortex. Proc Natl Acad Sci U S A. 2000;97(7):3661–5.
5. Carroll TJ, Barry B, Riek S, Carson RG. Resistance training enhances the stability of sensorimotor coordination. Proc Biol Sci. 2001;268(1464):221–7.
6. Carroll TJ, Riek S, Carson RG. Neural adaptations to resistance training—implications for movement control. Sports Med. 2001;31(12):829–40.
7. Carroll TJ, Selvanayagam VS, Riek S, Semmler JG. Neural adaptations to strength training: moving beyond transcranial magnetic stimulation and reflex studies. Acta Physiol. 2011;202(2):119–40.
8. Classen J, Liepert J, Wise SP, Hallett M, Cohen LG. Rapid plasticity of human cortical movement representation induced by practice. J Neurophysiol. 1998;79(2):1117–23.
9. de Rugy A, Davoodi R, Carroll TJ. Changes in wrist muscle activity with forearm posture: implications for the study of sensorimotor transformations. J Neurophysiol. 2012;108(11):2884–95.
10. Diedrichsen J, White O, Newman D, Lally N. Use-dependent and error-based learning of motor behaviors. J Neurosci. 2010;30(15):5159–66.
11. Evarts EV. Relation of pyramidal tract activity to force exerted during voluntary movement. J Neurophysiol. 1968;31(1):14–27
12. Fetz EE, Cheney PD. Postspike facilitation of forelimb muscle-activity by primate corticomotoneuronal cells. J Neurophysiol. 1980;44(4):751–72.
13. Georgopoulos AP, Kalaska JF, Caminiti R, Massey JT. On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. J Neurosci. 1982;2(11):1527–37.
14. Giesebrecht S, van Duinen H, Todd G, Gandevia SC, Taylor JL. Training in a ballistic task but not a visuomotor task increases responses to stimulation of human corticospinal axons. J Neurophysiol. 2012;107(9):2485–92.
15. Hamel-Paquet C, Sergio LE, Kalaska JF. Parietal area 5 activity does not reflect the differential time-course of motor output kinetics during arm-reaching and isometric-force tasks. J Neurophysiol. 2006;95(6):3353–70.
16. Hammerbeck U, Yousif N, Greenwood R, Rothwell JC, Diedrichsen J. Movement speed is biased by prior experience. J Neurophysiol. 2014;111(1):128–34.
17. Hortobagyi T, Richardson SP, Lomarev M, et al. Chronic low-frequency rTMS of primary motor cortex diminishes exercise training-induced gains in maximal voluntary force in humans. J Appl Physiol. 2009;106(2):403–11.
18. Huang VS, Haith A, Mazzoni P, Krakauer JW. Rethinking motor learning and savings in adaptation paradigms: model-free memory for successful actions combines with internal models. Neuron. 2011;70(4):787–801.
19. Hutton RS, Enoka RM, Suzuki S. Activation history and constant errors in human force production. Brain Res. 1984;307(1–2):344–6.
20. Kakei S, Hoffman DS, Strick PL. Muscle and movement representations in the primary motor cortex. Science. 1999;285(5436):2136–9.
21. Kakei S, Hoffman DS, Strick PL. Direction of action is represented in the ventral premotor cortex. Nat Neurosci. 2001;4(10):1020–5.
22. Kleim JA, Hogg TM, VandenBerg PM, Cooper NR, Bruneau R, Remple M. Cortical synaptogenesis and motor map reorganization occur during late, but not early, phase of motor skill learning. J Neurosci. 2004;24(3):628–33.
23. Krakauer JW, Pine ZM, Ghilardi MF, Ghez C. Learning of visuomotor transformations for vectorial planning of reaching trajectories. J Neurosci. 2000;20(23):8916–24.
24. Larrabee MG, Bronk DW. Prolonged facilitation of synaptic excitation in sympathetic ganglia. J Neurophysiol. 1947;10(2):139–54.
25. Lee M, Gandevia SC, Carroll TJ. Short-term strength training does not change cortical voluntary activation. Med Sci Sports Exerc. 2009;41(7):1452–60.
26. Muellbacher W, Ziemann U, Wissel J, et al. Early consolidation in human primary motor cortex. Nature. 2002;415(6872):640–4.
27. Nuzzo JL, Barry BK, Gandevia SC, Taylor JL. Acute strength training increases responses to stimulation of corticospinal axons. Med Sci Sports Exerc. 2016;48(1):139–50.
28. Orban de Xivry JJ, Criscimagna-Hemminger SE, Shadmehr R. Contributions of the motor cortex to adaptive control of reaching depend on the perturbation schedule. Cereb Cortex. 2011;21(7):1475–84.
29. Rosenkranz K, Kacar A, Rothwell JC. Differential modulation of motor cortical plasticity and excitability in early and late phases of human motor learning. J Neurosci. 2007;27(44):12058–66.
30. Selvanayagam VS, Riek S, Carroll TJ. Early neural responses to strength training. J Appl Physiol. 2011;111(2):367–75.
31. Shadmehr R, Mussaivaldi FA. Adaptive representation of dynamics during learning of a motor task. J Neurosci. 1994;14(5):3208–24.
32. Slijper H, Richter J, Over E, Smeets J, Frens M. Statistics predict kinematics of hand movements during everyday activity. J Mot Behav. 2009;41(1):3–9.
33. Smyrnis N, Mantas A, Evdokimidis I. “Motor oblique effect”: perceptual direction discrimination and pointing to memorized visual targets share the same preference for cardinal orientations. J Neurophysiol. 2007;97(2):1068–77.
34. Stefan K, Kunesch E, Cohen LG, Benecke R, Classen J. Induction of plasticity in the human motor cortex by paired associative stimulation. Brain. 2000;123:572–84.
35. Tanaka H, Sejnowski TJ, Krakauer JW. Adaptation to visuomotor rotation through interaction between posterior parietal and motor cortical areas. J Neurophysiol. 2009;102(5):2921–32.
36. Taube W. “What trains together, gains together”: strength training strengthens not only muscles but also neural networks. J Appl Physiol. 2011;111(2):347–8.
37. Verstynen T, Sabes PN. How each movement changes the next: an experimental and theoretical study of fast adaptive priors in reaching. J Neurosci. 2011;31(27):10050–9.
38. Yanai Y, Adamit N, Israel Z, Harel R, Prut Y. Coordinate transformation is first completed downstream of primary motor cortex. J Neurosci. 2008;28(7):1728–32.
39. Ziemann U, Iliac TV, Pauli C, Meintzschel F, Ruge D. Learning modifies subsequent induction of long-term potentiation-like and long-term depression-like plasticity in human motor cortex. J Neurosci. 2004;24(7):1666–72.
40. Ziemann U, Muellbacher W, Hallett M, Cohen LG. Modulation of practice-dependent plasticity in human motor cortex. Brain. 2001;124:1171–81.


© 2016 American College of Sports Medicine