|Home | About | Journals | Submit | Contact Us | Français|
When grasping and lifting different objects, visual cues and previously acquired knowledge enable us to prepare the upcoming grasp by scaling the fingertip forces according to the actual weight of the object. However, when no visual information is available, the object’s weight has to be predicted based on information learned from previous grasps. Here, we investigated changes in corticospinal excitability (CSE) and grip force scaling depending on the presence of visual cues and the weight of previously lifted objects. CSE was assessed by delivering transcranial magnetic stimulation (TMS) at different times before grasp of the object. In conditions where visual information was not provided, the size of motor evoked potentials (MEP) was larger when the object lifted was preceded by a heavy relative to a light object. Interestingly, the previous lift also affected MEP amplitude when visual cues about object weight were available, but only in the period immediately after (50 ms) object presentation; this effect had already declined for TMS delivered 150 ms after presentation. In a second experiment, we demonstrated that these CSE changes are used by the motor system to scale grip force. This suggests that the corticospinal system stores a ‘sensorimotor memory’ of the grasp of different objects and relies on this memory when no visual cues are present. Moreover, visual information about weight interacts with this stored representation and allows the corticospinal system to switch rapidly to a different model of predictive grasp control.
Initiation of grasp involves the coordination of both external sensory inputs and internal models of the desired movement. Sensory information related to the size, shape and orientation of objects is provided through vision (Gordon et al., 1991; Jenmalm and Johansson, 1997). Internal models related to object weight are learned through previous motor experience and allow prediction of the actual grip force appropriate for the object load (Flanagan and Wing, 1997; Witney et al., 2000; Wolpert and Flanagan, 2001).
When lifting an object, we use a force that is precisely scaled to its weight. This accurate scaling of the lift force relies on predictions based upon previous lifts or estimation of the weight of a particular object based on available visual information. When presented with an object of unknown weight, subjects will first use an inappropriate force, which is usually an overestimate of the actual force required to lift the object. Subjects adjust and refine the force needed during subsequent lifts of the same object to achieve the optimal force required. This rapid adaptation of force generation indicates that there is a real-time motor updating of the information used to coordinate the grasp (Johansson and Westling, 1984, 1988).
Johansson and Westling (1988) first showed that information acquired while subjects grasped and lifted an object influenced the preparatory grip force used for a subsequent lift. More recently, Chouinard et al. (2005) found that the primary motor (M1) and dorsal premotor cortices played an important role in force scaling depending on the previous lift and on visual cues, respectively (Chouinard et al., 2005). Although M1 seems to be involved in storing, or recruiting from other areas, a ‘sensorimotor memory’ about previous lifts, it is not known how this ‘memory’ is represented in M1 and how visual information about weight, when it becomes available, interacts with this ‘memory’. Therefore, we investigated whether corticospinal excitability is influenced by the sensorimotor memory about the previous lift and how and when visual information about an object’s weight interacts with this memory of the previous lift.
The rationale for the present study was to examine CSE and grip-lift performance for two pairs of drinking glasses: for one pair, the glass was transparent providing a clear visual cue to whether the glass was full (heavy) or nearly empty (light); the other pair of glasses were identical, except that the glass was opaque and therefore its weight was unknown to the subject. We sought to analyse (i) the simple effect of sensorimotor memory by comparing the MEPs obtained when preparing to lift an object of unknown weight preceded either by a heavy or by a light object; (ii) the simple effect of visual cues by comparing lifts of visibly heavy vs visibly light objects, both preceded by lift of an object having the same weight; and (iii) the interactions between visual input and sensorimotor memory by comparing trials of visible objects preceded by trials in which heavy or light objects were lifted.
We recruited 8 subjects for experiment #1 (5 males, 24±4 years) and 7 for experiment #2 (3 males, 23±5 years; 5 out of experiment #1). All subjects were right-handed according to the Edinburgh handedness inventory (Oldfield, 1971). Their vision was normal, or corrected to normal, and none had a history of neurological disease. Subjects were screened for potential risk of adverse reactions to TMS by using the transcranial magnetic stimulation adult safety screen (Keel et al., 2001). All subjects gave their written informed consent.
Subjects were seated in a relaxed position with their right hand placed on a hand-pad 30 cm away from a carousel device (Fig. 1A). This device could present one of 4 different plastic drinking glasses of identical shape (5 g weight, 50 mm diameter and 10 cm tall) but containing different weights (2 × ‘light’: 5g; 2 × ‘heavy’: 300g). One pair of glasses was opaque so that subjects could not deduce their weight, the others were transparent. The carousel was viewed through a screen, which was either opaque or transparent. The screen was placed in front of the subject to block the subject’s view while the carousel rotated between trials to present, in pseudorandom order, the next glass (see (Davare et al., 2009); Davare et al., 2010). Once the carousel stopped and the screen turned transparent (‘Go’ signal), the subjects had to initiate, at their own pace, a precision grip of the glass using only the right thumb and the index finger and to lift it a few cm above the carousel. The room was kept dark except for a small headlamp that illuminated the carousel and the objects. This was to ensure that subjects were not distracted or affected by any other external cues.
The objects were presented randomly in 6 blocks of 40 trials. Each particular condition (tms timing × object; see below) was repeated 20 times.
We used a custom-made figure-of-eight coil (7 cm outer diameter) connected to a single-pulse monophasic Magstim model 200 stimulator (Magstim Company, Whitland, UK). The stimulus was delivered over M1, with posterior to anterior induced current, through a coil held perpendicularly to the central sulcus with the handle pointing backwards. To target M1, the coil was positioned over the left motor cortex at the site where the MEP amplitude was the greatest in both first dorsal interosseous (1DI) and abductor pollicis brevis (APB) muscles. A single TMS pulse (120% of the resting motor threshold (Rossini et al., 1994)) was delivered at different times during preparation for grasp, but before grasp itself, i.e. at 50 ms, 100 or 150 ms after object presentation. The average resting motor threshold of all the subjects was 40 ± 6% of the maximal stimulator output.
The Magstim stimulator was triggered using Spike2 software and a Power 1401 CED data acquisition interface (Cambridge Electronic Design, Cambridge, UK). Electromyographic (EMG) activity was recorded with bipolar surface electrodes (belly-tendon), one pair positioned over the 1DI and the other over APB. The raw EMG signals were amplified (1K) (Neurolog, Digitimer Ltd, UK) and digitized at 5 kHz for offline analysis.
The amplitude of the evoked MEP was measured to assess the changes in CSE during the preparation of the different grasps. To normalise the MEP data across subjects, we computed the ratio between MEPs gathered during preparation of lifts preceded by a heavy and light object (MEPheavy/MEPlight); a value above 1 indicating a facilitatory effect when a heavier object preceded the lift. In addition, to assess the pure effect of visual cues, we computed the ratio between MEPs recorded during preparation of lift of a visibly heavy and a visibly light object, but both preceded by an object of the same weight to cancel out the sensorimotor memory effect. A value above 1 indicating an MEP facilitation when “heavy” visual cues were presented.
Subjects were seated comfortably with their right hand placed in a relaxed position on a table. The subjects were required to grip and lift a 225 g manipulandum with only their thumb and index finger (Fig. 1B). The manipulandum consisted of two parallel vertical grip surfaces of smooth aluminium (40 mm diameter, 30 mm apart, see (Davare et al., 2006) for details). The grip surfaces covered three-dimensional force-torque sensors (Mini 40 F/T transducers; ATI Industrial Automation, Garner, NC). Each sensor measured the three orthogonal forces (Fx, Fy, Fz) along the corresponding axes intersecting the centre of the grip surface. The lift force tangential to the grip surface (load force, LF) was given by Fy. The force normal to the grip surface (grip force, GF) was given by Fz. A hole was drilled into the table beneath the manipulandum to contain a hollow steel cylinder (Fig. 1B). A load of 300 g was placed inside the cylinder and could be attached to the manipulandum to increase its weight without the subject being aware that the load had been changed until they lifted the manipulandum. TMS was applied every 10 s (jitter 1.5 s) while subjects were with their hand at rest, 2 cm in front of the manipulandum. TMS was used as the ‘GO’ signal. Subjects had to lift the manipulandum to a height of 5 cm at their own pace and replace it back to its original position after 2 s. Talcum powder was applied to the subjects’ thumb and index finger in order to keep the friction between the grip surfaces and fingers constant.
Subjects performed 4 blocks of 21 trials. In half of the blocks, TMS was applied as a sham, by placing the coil orthogonal to the scalp, to control for any disruptive effect of TMS on the grip force scaling. Within blocks, the 21 trials were presented in a pseudorandom order to have 5 of each transition between weights, i.e. light-after-light, heavy-after-light, heavy-after-heavy, light-after-heavy. Overall, each weight transition was repeated 10 times.
The GF and LF of the thumb and index finger were digitized using the CED Power 1401 interface. For each trial, we measured the peak GF rate, given by computing the first derivative of GF using Matlab (Natick, US). Surface electrodes were also placed at subject’s 1DI and APB to measure the MEPs and EMG activity during each trial. Signal 3 (CED, Cambridge, UK) was used to record and measure the MEPs and fingertip forces during each trial.
In experiment #1, the actual MEP amplitudes were analysed by using 2-way repeated measure ANOVAs (ANOVARM), performed separately for the objects of visible weight (transparent objects) and unknown weight (opaque objects). Within-subject factors were tms timing (50, 100 or 150 ms after object presentation) and previous lift weight (light or heavy). For the objects of visible weight, an additional within-subject factor was the actual visible object weight (light or heavy). Corrections for violation of sphericity were done using the Greenhouse-Geisser correction. Linear regressions were used to investigate the evolution of the CSE during the 3 tms timings. In experiment #2, the MEP amplitudes and the GF scaling were analysed using two-way ANOVARM with previous lift weight (light or heavy) and tms (real or sham) as within-subject factor. Post-hoc analyses were performed using Tukey’s tests.
The aim of experiment #1 was to investigate whether lifting objects of different weights changed the CSE. TMS was used to assess CSE at three different times before the actual grasp, i.e. 50, 100 or 150 ms after the object was presented.
To understand how the sensorimotor memory of the previous lift affects the CSE, we first analysed trials where TMS was delivered during preparation of a lift of an object of unknown weight (opaque objects either light or heavy; Fig. 2A). The ANOVARM performed on the 1DI and APB MEP amplitudes showed a main effect of the previous lift weight (both F>7.84, both p<0.029) and no main effect of tms timing (both F<1). The MEP amplitudes were significantly larger when the object presented was preceded by a heavy one (mean 40%; all p<0.034). Since there was no effect of tms timings on the MEP amplitude, this could indicate that the influence of sensorimotor memory from the previous trial remained constant across preparation of the upcoming lift of an unknown weight.
In order to understand how visual information about the object’s weight interacts with the sensorimotor memory, we analysed trials where TMS was applied during preparation of lifts of objects which have their content clearly presented to subjects enabling them to determine if it is light or heavy. Interestingly, we still found an effect of the weight of the previous lift on the MEP size when the upcoming object weight was cued by vision, but this effect was restricted to TMS applied 50 ms after object presentation (Fig. 2B). Indeed, the ANOVARM showed an interaction between tms timing and previous lift weight (ANOVARM on FDI and APB MEPs: both F>5.31, both p<0.021). Post-hoc tests disclosed that the 1DI and APB MEP amplitudes were larger when the previous lift was heavy and when TMS was applied 50 ms after object presentation (all p<0.031). However, when TMS was applied 100 and 150 ms after object presentation, the sensorimotor memory effect of the previous lift was suppressed (all p>0.137; Fig. 2B). There was a significant difference of the previous weight effect between the 50 and 150 ms tms timings (both p<0.017), but not between the 50 and 100 ms tms timings (both p>0.05). However, a linear regression performed with each value (50, 100 and 150 ms tms timings) of all 8 subjects showed a significant negative slope between the MEP amplitudes and the 3 tms timings (both p<0.003).
This suggests that the visual information about an object’s weight interacts with the sensorimotor memory within M1 as soon as 100 ms after being presented. By 150 ms after information cueing the correct weight of an object is made available, the motor system has made a rapid switch from the previously used internal predictive model to the one cued by available sensory information.
In order to assess the simple effect of visual weight cues, we analysed lifts performed on either visible heavy or light objects, but both preceded by a lift of objects having the same weight to cancel out the sensorimotor memory effect. We found that visual information strongly facilitated CSE 150 ms after object presentation when a heavy object was presented compared to a light (all p<0.007; Fig. 2C). ‘Heavy’ visual cues did not significantly increase CSE 100 ms after object presentation; however, a linear regression showed a significant positive slope between the MEP sizes gathered at each tms timing (both p<0.008). This indicates that visual information about weight influences the CSE, but only after it becomes available to the corticospinal system.
Experiment #2 was designed to test whether the increased CSE when lifts were preceded by heavier objects actually corresponded to the planning of a higher grip force scaling. To do so, subjects were required to lift a manipulandum without any external cue about its weight. The weight was changed in a pseudorandom order within a block without the subjects’ knowledge.
When the current lift was preceded by the heavier weight, the GF was applied at a higher rate than if it was preceded by the same light object (see arrows* Fig. 3A). Regarding the effect of a lighter weight in the previous lift, the force generation followed a course similar to that of the foregoing lift (low GF rate, see arrows*, Fig. 3B) until the point where the object would have started to move with the previous weight. Because of the absence of movement, forces continued to increase (see arrow** Fig. 3B), but still at a low rate, until the load was overcome (Fig. 3B). This is consistent with earlier findings (Johansson and Westling, 1988).
To analyse these effects, which involved four possible weight transitions (light-after-light, heavy-after-light, heavy-after-heavy, light-after-heavy), we computed two different ratios to describe how subjects scaled their GF and CSE depending on the sensorimotor memory. The first ratio is given by dividing the values of GF and CSE in light-after-heavy trials by the values in light-after-light trials; a ratio above 1 indicated a facilitatory effect of a heavy preceding a light object, compared to a light object. The second ratio is calculated by dividing GF and CSE values in heavy-after-heavy trials by the values in heavy-after-light trials; here a ratio below 1 signifies an inhibitory effect of a light preceding a heavy object, compared to a heavy object.
When analysing the GF rate, the performance in the 4 different weight transitions was consistent across all 7 subjects. There was a 20% increase in the GF rate for light-after-heavy compared to light-after-light (ratio = 1.21±0.12, mean±SD, n=7; p=0.008); and a 20% decrease in the GF rate for heavy-after-light compared to heavy-after-heavy (ratio = 0.78±0.09, mean±SD, n=7; p=0.024). This finding corroborates results from previous studies (Johansson and Westling, 1988). It is noteworthy that delivering TMS had no effect on the grip force scaling (ANOVARM, tms main effect (real vs sham), F<1).
In addition, we also looked at the CSE during the preparation of the subsequent lift (Fig. 4A). The CSE data replicated the results of experiment #1 and paralleled the results of the GF rate. The mean MEP amplitude ratio showed an increase in the MEP size for light-after-heavy lifts compared to light-after-light (p=0.003); and a decrease in MEP amplitude for heavy-after-light lifts compared to heavy-after-heavy (p=0.012). Interestingly, we found a correlation between the GF rate ratio and the MEP facilitation in both muscles (Fig. 4B-C only shows 1DI values; light object: r=0.74, p=0.004; heavy object: r=0.82, p=0.001). This might indicate that the more a previous heavier lift had a facilitatory effect on the MEP size during movement preparation, the more the GF rate reached higher levels in the subsequent lift.
The present study was designed to investigate the presence of a ‘sensorimotor memory’ effect on the corticospinal system when lifting objects of different weight. In addition, we sought to determine how visual information, when it becomes available, interacted with this sensorimotor memory. To do so, we analysed (i) the simple effect of sensorimotor memory by comparing the MEPs of lifts of an object of unknown weight preceded either by a heavier or lighter object, (ii) the simple effect of visual cues by comparing lifts of visible heavy vs light objects, both preceded by lift of an object having the same weight in order to cancel the memory effect, and (iii) the interactions between visual cues and sensorimotor memory by comparing lifts of visible objects that were preceded by either a heavy or light weight.
Our results clearly demonstrate that, when no external input about an object’s weight is available, the CSE is scaled according to the weight of the object experienced in the previous lift. Interestingly, when preparing the lift of an object of visible weight, the sensorimotor memory effect was still present early after object presentation (50 ms), but was gradually suppressed at 100 ms and completely cancelled by 150 ms after object presentation, as soon as visual information reached the corticospinal system. By comparing lifts with a similar ‘sensorimotor history’, we confirmed that the effect of visual cues only occurred late (150 ms) after object presentation. In a second experiment, we provided evidence that the facilitation in the MEP size when preparing a lift preceded by a heavier weight was correlated with a higher grip force rate subjects actually used to lift the object. This demonstrates that the corticospinal system is able to store an internal representation of motor outputs related to weight and that, as soon as visual cues are available, it can rapidly adapt its state to generate the most appropriate motor command and this is reflected in the excitability level of the relevant muscle representations.
Previous studies have shown that M1 is part of a larger cortical network involved in storing a sensorimotor memory of the previous lift. Functional imaging studies indicate that a large network including fronto-parietal cortical areas as well as the cerebellum is active when grasping and lifting objects of different weight (Schmitz et al., 2005; Jenmalm et al., 2006). Disrupting M1 function by repetitive TMS induces deficits in accurately scaling the GF in subsequent lifts when lifting objects of the same weight (Nowak et al., 2005) or in scaling the GF according to the preceding weight (Chouinard et al., 2005; Berner et al., 2007). Indeed, on the one hand, when objects had the same weight, delivering repetitive TMS over M1 led to an overshoot in the GF (Nowak et al., 2005). On the other hand, when lifting objects of different weights, GF is scaled according to the previous lift based on a trial-by-trial sensorimotor memory, that is GF is either too low when a lighter weight was expected, or too high when a heavier weight was expected (Johansson and Westling, 1988). Chouinard et al. (2005) found that, following repetitive TMS, their subjects lost the ability to scale grip forces according to the immediately preceding lift. Although these studies provide a causal relationship between the processing taking place in M1 and the establishment of a sensorimotor memory allowing the precise scaling of the grip force, here we further demonstrate that M1 could store this sensorimotor memory by modulating CSE of the involved muscle representation; a higher CSE level giving rise to a higher grip force rate (Fig. 4). There is evidence from previous studies of a sensorimotor memory in M1 (Li et al., 2001; Nowak et al., 2005; Berner et al., 2007), and it is therefore plausible that the MEP changes we have documented might have originated from M1 itself. However, we cannot exclude the possibility that other descending influences may have altered levels of spinal motoneuronal excitability independent of M1 input. The cerebellum, another likely source of internal models for predictive grasp (Schmitz et al., 2005; Bursztyn et al., 2006; Jenmalm et al., 2006), may exert much of its influence through motor cortex and therefore contribute to M1 excitability changes.
In addition, we found that the motor system can make a rapid switch from a previously used internal predictive model to a different one by available sensory information. In order to explain how CSE is influenced by the previous weight lifted and by visual cues, we formulated a model of interactions between the stored sensorimotor memory and available visual information (Fig. 5). The upper part shows schematically the raw MEP size for each object at the three TMS timings tested. At 50 ms after object presentation, the CSE is only influenced by the previous weight because only the ratios between objects of different weight history are above 1 (b/a and d/c; Fig. 5). Indeed, ratios between objects of similar weight history and different visible weight showed values of 1. In contrast, 150 ms after object presentation, the ratios reversed and now have a ratio above 1 between objects of different visible weight (c/a and d/b; Fig. 5) but not for objects of different weight history. Transitional results were found 100 ms after visual cues are presented. It is noteworthy that, when preparing the lift of a visible light object that was preceded by a heavier one, the effect of the ‘light’ visual cue decreased CSE. Conversely, preparing the lift of a heavy object preceded by a lighter one yielded the ‘heavy’ visual information to increase CSE. The evolution of CSE is depicted in the lower part of Figure 5. As soon as visual cues are available to the motor system, the CSE gradually moves from its anterior state, defined by the previous weight, to a state that corresponds to the visual information. The fact that visual information only interacts with CSE after 100 to 150 ms is in agreement with previous studies. Visual cues for movement reach human premotor areas at around 100 ms (Terao et al., 1998; Schluter et al., 1999). In a recent study, Prabhu et al. (Prabhu et al., 2007) reported that visual object cues could not influence CSE before 150 ms after object presentation. In monkeys, it has been reported that, on a visual conditional task, the mean onset of signal-related units in the premotor cortex was approximately 140 ms after the instruction stimuli (Johnson et al., 1996).
A predictive scaling of grip force adapted to an object’s weight can be explained within the theoretical framework of internal forward models (Wolpert and Flanagan, 2001). Indeed, when lifting an object, the subject’s own upward movement causes the load force to increase and, to prevent slippage, the grip force has to be increased in a parallel fashion (Flanagan and Wing, 1997). Due to the inevitable delays in cutaneous afferent pathways (Johansson and Westling, 1984), this predictive modulation of the grip force cannot be based upon peripheral feedback. Therefore, anticipatory grip force increases must be generated by using a predictive model of the consequences of the action (Flanagan and Wing, 1997; Blakemore et al., 1998). Our results suggest that this predictive internal model sets a sensorimotor memory in M1 by modulating the CSE. Interestingly, Quaney et al. (2003) found that this sensorimotor memory was not specific for object lifting; which suggests that previous actions bias M1 outputs even if they are not related to the present action. This could be a more ecological way for the motor system to learn actions in a stochastic environment (Quaney et al., 2003). In our present experiment, one strategy could be to generate a force midway between the two possible weights. However, this would require the motor system to rescale the motor output too often. Therefore, we suggest that a strategy based upon the previous action, as used by subjects in the present experiment, is more optimal.
In conclusion, we provide evidence that M1 stores a sensorimotor memory of an object’s weight by changing the level of excitability of the involved muscle representations. If visual information becomes available, it is then used rapidly to switch between different models of predictive grasp control.
MNL received a Wellcome Trust scholarship. MD is funded by the Wellcome Trust (WT083450) and the FRS-FNRS (Belgium). We thank Ed Bye and Victor Baller for the design of the turntable and manipulandum.