Anticipatory synergy adjustments: Preparing a quick action in an unknown direction

Abstract

We studied a mechanism of feed-forward control of a multi-finger action, namely anticipatory synergy adjustments (ASAs) prior to a quick force correction in response to a change in the gain of the visual feedback. Synergies were defined as co-varied across trials adjustments of commands to fingers that stabilized (decreased variance of) the total force. We hypothesized that ASAs would be highly sensitive to prior information about the timing of the action but not to information on its direction, i.e., on whether the gain would go up or down. The subjects produced accurate constant total force by pressing with four fingers on individual force sensors. The feedback signal could change from veridical (the sum of finger forces) to modified, with the middle finger force multiplied by 0.2 or by 1.8. The timing of the gain change and its direction could be known or unknown to the subject in advance. When the timing of the gain change was known, ASA was seen as a drop in the synergy index starting about 250–300 ms prior to the first visible correction of the total force. When the gain change timing was unknown, ASAs started much later, less than 100 ms prior to the total force correction. The magnitude of synergy index changes was significantly larger under the “time known” conditions. Information on the direction of the visual gain change had no effect on the ASA timing, while the ASA magnitude was somewhat larger when this information was not available to the subject. After the total force correction, the synergy index was significantly larger for the force signal computed using the modified gain values as compared to the synergy index value for the actual total force. We conclude that ASAs represent an important feed-forward motor control mechanism that allows preparing for a quick action even when the direction of the action is not known in advance. The results emphasize the subtle control of multi-finger synergies that are specific to the exact contributions of individual fingers to performance variables. The data fit well the central back-coupling hypothesis of synergies and the idea of control with referent body configurations.

Introduction

When a person gets ready to perform a quick action that explicitly requires a change in a certain mechanical variable, two types of anticipatory adjustments can be seen that are not resulting in changes in that variable. The first group involves changes in muscle activation that do not have effects on the variable to be changed, but they change other variables related, for example, to a postural task component. Typical examples are anticipatory postural adjustments (APAs, reviewed in Massion 1992), which represent changes in the activation levels of trunk and leg muscles that are seen prior to the initiation of an instructed action (for example, quick arm movement, Belenkiy et al. 1967) or prior to an expected perturbation (as during catching a load, Shiratori and Latash 2001). Their purpose has been assumed to mitigate or cancel postural disturbances associated with the planned action and/or an expected perturbation. APAs can be seen in time profiles of muscle activations and forces/moments averaged across repetitive attempts at the same task.

The second group of anticipatory adjustments is intimately related to the notion of multi-element synergies developed within the principle of motor abundance (Gelfand and Latash 1998; Latash 2012). Synergies have been defined as neural organizations that ensure co-variation of variables produced by elements of an apparently redundant system (elemental variables) that stabilizes a value or a time profile of a potentially important performance variable (Latash 2008). Synergies have been quantified using the computational apparatus of the uncontrolled manifold (UCM) hypothesis (Scholz and Schöner 1999; reviewed in Latash et al. 2007). This method involves quantifying two components of variance within the space of elemental variables across tasks. The first component (within the UCM, V_UCM) leads to no changes in the potentially important performance variable, while the second one (orthogonal to the UCM, V_ORT) does. An index of synergy (ΔV) has been used reflecting the normalized difference between V_UCM and V_ORT, both quantified per degree-of-freedom in the corresponding spaces.

Recently, several studies have provided evidence that the synergy index shows a drop about 200–250 ms prior to a self-paced action (Olafsdottir et al. 2005; Shim et al. 2005; Klous et al. 2012) or prior to a predictable perturbation (Kim et al. 2007; Krishnan et al. 2012). These phenomena have been termed anticipatory synergy adjustments (ASAs). Their purpose has been assumed to attenuate synergies stabilizing a variable in anticipation of its quick change. Note that during ASAs, there are changes only in co-variation patterns among elemental variables, while averaged across trials data show no changes in any variables.

The hypothesized functions of the two groups of anticipatory adjustments allow making predictions with respect to their changes when only part of the information about an upcoming action (perturbation) is known in advance. For example, if a person knows the exact timing of the action (perturbation) but not its direction, using APAs may be detrimental to postural stability. Indeed, if force/moment changes associated with APAs are in a wrong direction, they will exacerbate the effects of the perturbation, not attenuate them. In contrast, ASAs facilitate actions in any direction. They are not associated with changes in any performance variables and, as a result, they do not make direct effects of a perturbation stronger or weaker. They facilitate an action (a response to a perturbation) independently of its direction.

The notion of ASAs may be considered within a hypothesis that the production of an action by a redundant system involves changes in neural variables of two types (Latash et al. 2005). Neural variables of the first type produce changes in performance variables, to which all the elements contribute; these may be components of an explicitly planned action or of APAs. Neural variables of the second type define stability properties of those performance variables reflected in the corresponding synergy indices. ASAs are consequences of changes in the latter group of neural variables. As a result, when there is no advance knowledge about the direction of a planned action but the timing of the action is known, ASAs may be expected while APAs are not.

In this study, we used software-generated “perturbations” in the total force produced by a redundant set of fingers. No change in the task mechanics happened, and the “perturbation” was produced by an instantaneous change in the way visual feedback signal was computed (one finger could be made artificially weaker or stronger). Our prediction was that ASAs would be seen prior to such “perturbations” if their timing were known in advance. We expected to see no (or significantly reduced) ASAs in conditions, when the subject would not be able to predict the timing of such “perturbations”. Our second prediction was that ASAs would be equally strong in conditions when the direction of a “perturbation” was known in advance and when it was unknown.

Methods

Participants

Eight healthy males [age: 26.5 ± 3.0 (SD) yr, mass: 77.2 ± 10.5 kg, height: 1.75 ± 0.05 m] participated in the experiment. All the participants had no history of neurological or motor disorders; they were all right-hand dominant with respect to hand usage during eating and writing. None of the participants had a history of long-term involvement in hand or finger activities such as typing and playing musical instruments. All participants gave informed consent according to the procedures approved by the Office for the Research Protections of the Pennsylvania State University.

Apparatus

Four six-component ATI force sensors (Nano-17, ATI Industrial Automation, Garner, NC) were used to measure vertical forces produced by the fingers. The sensors were attached to a customized flat panel. Each sensor was covered with sandpaper (300 grit) to increase the friction between the fingertips and the top surface of the sensors. Subjects sat in a chair facing the 19″ computer monitor, which showed real-time force feedback (Figure 1A). The right upper arm was placed into the wrist-forearm brace. Velcro straps were used to prevent forearm and wrist movement during tests. A smoothly shaped wooden piece was placed underneath the subject’s palm to achieve a comfortable hand posture, while the forearm rested on a foam support. The four force signals were digitized at 100 Hz with a 32-bit resolution with a customized LabView program (LabView 2011, National Instruments).

FIGURE 1.

(A): An illustration of the hand position. The subjects placed their palm on a wooden piece, and the forearm was held stationary with Velcro straps. The sensors were attached to a wooden frame. (B): Feedback during the single-finger ramp tasks. (C): The visual gain change task. The vertical dashed line shows the perturbation time. The black bars schematically showed the direction of the gain change for the target finger force. The subjects did not know the exact scaling coefficient, and the actual size of the bar was not proportional to the gain.

Experimental Procedures

The subjects performed all the tasks with the right hand only. Before each task, subjects were given an instruction and a demonstration by an experimenter, and they had 5 practice trials, enough to make them feel comfortable with the procedure. Before each trial, the subject was asked to place the fingertips on the sensor centers and relax the hand; the sensor signals were set at zero during that time. The experiment consisted of three blocks including 1) maximal voluntary contraction (MVC) tasks; 2) single-finger ramp tasks; and 3) visual gain change task.

MVC tasks

In the MVC tasks, subjects were instructed to press on the sensors with the four ngers together as hard as possible in a self-paced manner and achieve maximal total force level within 4 s. The subjects were instructed to relax immediately after reaching a maximal force. The sum of the nger forces was shown on the screen. The maximal total force (MVC_TOT) and the forces of individual ngers (MVC_i; i = {I, M, R and L} – index, middle, ring and little fingers, respectively) at the time of reaching MVC were measured and used to normalize the target force levels across subjects for the next tasks. The subjects performed two consecutive attempts, and the trial with the highest MVC value was used.

Single- nger ramp tasks

Subjects were required to press with one of the ngers and match with its force a template that consisted of a horizontal segment at 5% of the MVC_i of the task nger for the rst 4 s, followed by a slanted line from 5% to 40% of the MVC_i over the next 6 s, and a horizontal segment at 40% of MVC_i for the last 4 s (Figure 1B). Subjects were required to press with the task finger while keeping all the fingers on the sensors and to match the task finger force with the template shown on the screen. Subjects were instructed to pay no attention to possible force production by non-task ngers.

Visual gain change task

The subjects were required to press naturally with all four fingers and match the signal shown on the screen with a horizontal line corresponding to 20%MVC_TOT. The feedback signal (F_FB) shown on the monitor was computed as F_FB = F_I + k × F_M + F_R + F_L, where k is a gain factor used to manipulate the task feedback. Subjects knew in advance that only the gain at the middle finger force would be manipulated. Three values of k were used; k = 1 corresponded to veridical feedback signal equal to actual total force (F_TOT); k = 0.2 and k = 1.8 were used to either attenuate or increase the contribution of the middle finger force to F_FB. When a change in k happened, F_FB jumped up or down, and the subjects were required to adjust finger forces such that F_FB went back to the previous target level as quickly as possible.

Each episode involving a change in k (a “perturbation”) consisted of three intervals (Figure 1C). The bars on the screen schematically showed the direction of the gain change for the middle finger force. The subjects did not know the exact scaling coefficient, and the actual size of the bar was not proportional to the gain. For the first 3 s, k = 1. Over the next 4 s, a change in k could happen at any time. Subjects were asked not to change finger forces in anticipation of a change in k, and trials that showed signs of such anticipatory F_TOT changes were rejected and repeated. The final 3 s was the recovery time. At the beginning of this time interval, k changed back to k = 1. Subjects had 3 s to recover to the target force under the restored veridical feedback.

Each trial took 31 s; it involved 4 s of preparation time, during which the subjects were required to match the template as accurately as they could and 3 episodes with k changes. No recovery interval was given at the end of the last (third) episode.

Three factors were manipulated in the experiment: the direction of the k change, the prior knowledge on the k change timing, and the prior knowledge on the k change direction resulting in the 2 × 2 × 2 design. The direction of the k change was either from k = 1 to k = 0.2 or from k = 1 to k = 1.8. The timing of the k change could be known in advance (shown on the screen as a vertical dashed line) or unknown (it could happen at any time within the 4-s window). The direction of the k change could be known in advance (shown on the screen as in Figure 1C) or unknown. We will refer to the four conditions of the prior knowledge as T_KD_K (timing known, direction known), T_KD_U (timing known, direction unknown), T_UD_K (timing unknown, direction known), and T_UD_U (timing unknown, direction unknown). There were 16 trials for each condition; each trial contained 3 “perturbations” (visual gain changes). So for each condition, there were 48 perturbations. Half of those were from k = 1 to k = 1.8, while the other half were from k = 1 to k = 0.2. The conditions were randomized. The subjects had 2 min rest time after 10 trials.

Data analysis

A second-order zero-lag Butterworth filter with a cutoff frequency of 10 Hz was adopted to filter the data with Matlab (Matlab R2011a; The MathWorks). The filtered data were then processed off-line using a customized Labview program.

MVC task

Peak force (MVC) was measured at the time when the force produced by all four fingers peaked.

Single- nger ramp tasks

The enslaving 4×4 matrix (E) reflects the unintentional force productions by non-task fingers when an instructed finger produces force (Zatsiorsky et al. 1998, 2000). The E matrix was computed using the data from the single-finger ramp trials for each subject. For each single-finger trial, linear regressions of the forces produced by individual fingers against F_TOT over a 4-s time interval (starting 1 s after the ramp initiation and ending 1 s before the ramp termination to avoid edge effects) were computed. The regression coefficients (Eq. 1) were used to construct E (Eq. 2):

$$F_{i,j}=f_i^0+k_{i,j}•F_{\mathit{TOT},j}$$ (1)

$$E=\left[\left(\begin{array}{cccc}{k}_{I,I}& k_{I,M}& k_{I,R}& k_{I,L}\\ k_{M,I}& k_{M,M}& k_{M,R}& k_{M,L}\\ k_{R,I}& k_{R,M}& k_{R,R}& k_{R,L}\\ k_{L,I}& k_{L,M}& k_{L,R}& k_{L,L}\end{array}\right)\right]$$ (2)

Where i,j = {I,M,R,L}, j represents a task finger, F_i,j and F_TOT,j indicate the individual i-finger force and F_TOT,j, respectively. The enslaving matrix E, was used to convert the 4×1 force data vector f into a set of hypothetical commands to fingers (finger modes, Danion et al. 2003) by using the equation, m = E⁻¹ where m is the 4×1 mode vector.

Visual gain change task

For each episode with a change in k, the time of F_TOT change (t₀) was defined as the time when the first derivative of F_TOT (dF_TOT/dt) reached 5% of its peak value in that particular trial. Further, for all the episodes within each condition, time intervals starting 600 ms before t₀ and ending 1000 ms after t₀ were aligned by t₀ and averaged across episodes for each time sample. The trials with the finger force values that differed from the mean by more than 3 standard deviations were rejected.

Further analysis of multi-finger synergies stabilizing the actual total force profile (F_TOT) and the feedback force (F_FB) was performed within the framework of the UCM hypothesis (Scholz and Schöner 1999; reviewed in Latash et al. 2007). The hypothesis assumes that the controller organizes co-variation among elemental variables (finger modes) to stabilize a certain value of a performance variable (F_TOT or F_FB). The values of finger modes (m_i), were computed on the basis of the force magnitudes (F_i) and the matrix E for each time sample. After the trial alignment, variance in the m space across trials was quantified separately in two subspaces for each time sample. The four-dimensional space of finger modes can be divided into two subspaces, one corresponding to a fixed value of F_TOT or F_FB (the UCM, 3-dimensional space) and the other leading to changes in F_TOT or F_FB (orthogonal to the UCM, 1 dimensional). Variance across trials was computed for each time sample (each phase) and compared within the two subspaces, V_UCM and V_ORT after normalization per DOF. We interpret V_UCM > V_ORT as a sign of a multi-finger synergy stabilizing the total force.

A synergy index, ΔV was computed reflecting the relative amount of V_UCM in the total variance (V_TOT):

$$\mathrm{\Delta }V=({\text{V}}_{\text{UCM}}-{\text{V}}_{\text{ORT}})/{\text{V}}_{\text{TOT}},$$ (3)

where all variance indices are computed per degree of freedom. For statistical analysis, the ΔV values were transformed using a Fisher’s z-transformation adapted to the boundaries of ΔV:

$$\mathrm{\Delta }{V}_Z=0.5•log[(4+\mathrm{\Delta }V)/(1.33-\mathrm{\Delta }V)]$$ (4)

The averaged value and standard deviation (SD) of V_Z were computed for the steady-state (between −600 and −400 ms before t₀). Anticipatory synergy adjustments (ASA) were identified as a drop in the ΔV_Z time profile. The time of ASA initiation, t_ASA, was defined as the time when ΔV_Z dropped below its average steady-state value by more than 2SD. Negative value of t_ASA mean that ΔV_Z started to drop before the initiation of F_TOT changes. We also quantified the magnitude of the ΔV_Z drop (ΔΔV_Z) during the ASAs as the change in ΔV_Z from its steady-state value to the value at t₀.

Statistics

Data are presented in the text and figures as means and standard errors. The data were tested for normality using the Shapiro-Wilk test. Log-transformation was used for non-normally distributed data. Three-way ANOVA with repeated measures was used. In particular, we explored how the main outcome variables (such as ΔV_Z, ΔΔV_Z and t_ASA) were affected by factors Gain (two levels, from k = 1 to k = 1.8 and to k = 0.2), Time (two levels: known and unknown), and Direction (two levels: known and unknown). Before performing ANOVAs, ΔV were transformed using Fisher’s z-transformation, see Eq. (4). The statistical significance was set at p < 0.05. Statistical analysis was performed in SPSS 20 (SPSS, Inc., Chicago, USA)

Results

The subjects showed different time delays between the time of the change in the gain (k) and the first detectable change in the total force. Figure 2 shows an illustration of the total force (F_TOT, solid lines) and feedback force (F_FB, dashed lines) for a representative subject under two conditions, known (T_K, thin lines) and unknown (T_U, thick lines) “perturbation” time. At the time of perturbation (t_P), the gain changed from k = 1 to k = 0.2. This led to an instantaneous drop in F_FB without a change in F_TOT. F_TOT corrections were seen at a delay (adjustment time) that varied across conditions. It was, on average, the shortest when k increased from 1 to 1.8 and its timing was known (265.7 ± 20.7 ms) and longest when k dropped from 1 to 0.2 and its timing was unknown (401.4 ± 11.5 ms). The three-way repeated measures ANOVA with factors Gain, Time and Direction showed significant main effects of Gain [F_{(1, 7)} = 17.91; p < 0.05] and Time; [F_{(1, 7)} = 74.88; p < 0.01]. Post-hoc Scheffe tests confirmed the longer adjustment times for the gain change to 0.2 as compared to 1.8 and for the T_U condition as compared to T_K. There also was a significant Gain × Time interaction [F_{(1, 7)} = 6.19; p < 0.05]. It reflected the fact that the Time effect was more prominent in the condition of k increase from 1 to 1.8 as compared to k decrease from 1 to 0.2 There were no significant effects of Direction and no interactions with this factor.

FIGURE 2.

Typical total force (F_TOT, solid lines) and feedback force (F_FB, dashed lines) time profiles for a representative subject under the T_U (time unknown, thick lines) and T_K (time known, thin lines) conditions. The time of gain change (“perturbation time”) is shown as t_P. Note the longer adjustment time under the T_U condition.

Anticipatory Synergy Adjustments (ASAs)

During steady-state force production under veridical feedback, the subjects showed high indices of synergy (ΔV, see Methods) stabilizing total force under all conditions. There were no significant effects on ΔV_Z during the steady-state of any of the main factors (Gain, Time and Direction). Averaged across conditions and subjects ΔV_Z during the steady-state was 3.49 ± 0.055 (ΔV_Z > 0; t = 19.74, p < 0.01). Figure 3 illustrates time profiles of ΔV_Z computed across trials for a representative subject for two polar conditions, when both the time and direction of the gain change were known (T_KD_K), and when both the time and direction were unknown (T_UD_U). In this illustration the gain (k) changed from 1 to 0.2. The ΔV_Z index was computed with respect to actual total force (F_TOT; solid line, dark shade) and feedback force (F_FB, dashed line, lighter shade) shown on the screen. Since prior to the change in gain F_FB = F_TOT, ΔV_Z values were identical. After the change in k, F_FB was computed using the modified gain for the force of the middle finger. As a result, the analyses were performed with two Jacobians, [1,1,1,1] with respect to F_TOT and [1,0.2,1,1] with respect to F_FB.

FIGURE 3.

Averaged across subjects time profiles of the z-transformed synergy index (ΔV_Z) under the T_KD_K (time known directiion known, panel A) and T_UD_U (time unknown, direction unknown, panel B) conditions. Time zero (t₀) corresponds to the earliest change in the total force following a change in the gain (k). with repect to total force and feedback force. After t₀, the two lines represent ΔV_Z indices computed with respect to total force (Jacobian J = [1,1,1,1]; solid line, dark shade) and with respect to feedback force (Jacobian J = [1,0.2,1,1], dashed line, light shade). The shades show the standard error bars across subjects.

The following features of the ΔV_Z(t) curves can be noticed. First, there is an early drop in ΔV_Z starting about 300 ms prior to the change in finger force (time zero, t₀) under the T_KD_K condition. This drop represents the anticipatory synergy adjustment (ASA). ASA starts much later (less than 100 ms prior to t₀) and is much smaller in magnitude by t₀ under the T_UD_U condition. Second, there is a slow increase in ΔV_Z after the gain change, which is more pronounced for ΔV_Z computed with respect to F_FB as compared to that computed with respect to F_TOT.

We used two parameters to quantify ASAs, the time of ASA initiation as compared to t₀ (t_ASA) and the magnitude of ΔV_Z drop from the preceding steady-state to the magnitude at t₀ (ΔΔV_Z). Overall, when the subjects knew the time of the gain change (T_K), ASAs started earlier and were of a larger magnitude as compared to the T_U conditions. Knowing or not knowing the direction of the gain change in advance had no effect on either of the two parameters. These results are illustrated in Figure 4. Panel A illustrates ΔΔV_Z during ASA (averaged across subjects with standard error bars) while panel B shows the data for t_ASA. Overall, the subjects showed larger changes in ΔV_Z during ASAs when k dropped from 1 to 0.2 as compared with conditions when k increased from 1 to 1.8 (compare the black and white bars in Fig. 4A); there was no such a difference in the timing of ASAs (Fig. 4B).

FIGURE 4.

Indices if anticipatory synergy adjustments (ASAs). (A): The magnitude of the drop in the synergy index, ΔΔV_Z. (B): The time of initiation of the ASA (t_ASA) with respect to the earliest change in the total force following a change in the gain (k). Black bars show the data for the drop in k, while open bars show the data for the increase in k. T_K – time of the k change known; T_U – time unknown; D_K – direction known; D_U – direction unknown. Average values are presented with standard error bars.

Three-way repeated measures ANOVA on ΔΔV_Z with the factors Gain, Time and Direction showed significant main effects of all three factors [Gain: F_(1,7) = 10.11; p < 0.05; Time: F_(1,7) = 8.14; p < 0.05; Direction: F_(1,7) = 8.27; p < 0.05] with no interactions. A similar ANOVA on t_ASA showed a significant effect of Time [F_(1,7) = 70.54; p < 0.01] with no other effects.

We also asked a question whether, after a change in k, the subjects stabilized the total force (F_TOT) or the feedback showed on the screen (F_FB). To quantify the differences between the two analyses, we compared V indices computed using two different Jacobians corresponding to F_TOT and F_FB stabilization 1 s after the change in k. The difference (dΔV_Z) between the two z-transformed ΔV indices is presented in Figure 5 (averages across subjects with standard error bars). Note that all these values are consistently negative meaning that ΔV_Z computed with respect to F_FB was consistently higher than V_Z computed with respect to F_TOT. The difference dΔV_Z was consistently lower for the conditions with an increase in k from 1 to 1.8 as compared to conditions with k drop from 1 to 0.2. These effects were confirmed by a three-way repeated measures ANOVA on dΔV_Z with the factors Gain, Time and Direction, which showed a significant main effect of Gain [F_(1,7) = 38.08; p < 0.01] without other effects.

FIGURE 5.

The difference (dΔV_Z) between the synergy index values computed at 1 s after the k change with respect to the total force and with respect to the feedback force. Black bars show the data for the drop in k, while open bars show the data for the increase in k. T_K – time of the k change known; T_U – time unknown; D_K – direction known; D_U – direction unknown. Average values are presented with standard error bars.

Discussion

The experiment confirmed the two main predictions made in the Introduction. Indeed, anticipatory synergy adjustments (ASAs) were seen 250–300 ms prior to action initiation when the subjects knew in advance the exact timing of the “perturbation” (change in the gain factor k). In contrast, ASAs were seen much later (< 100 ms prior to the “perturbation”) and their magnitude was much smaller in conditions when the exact timing of the k change was unknown. The most important observation, however, is that the timing and magnitude of ASAs were about the same in conditions when the subjects knew in advance the direction of the “perturbation” (whether the gain would go up or down) and in conditions when they had no such knowledge. In fact, there were larger anticipatory changes in the synergy index in conditions when the direction of k change was unknown.

ASAs were discovered about eight years ago (Olafsdottir et al. 2005; Shim et al. 2005) in studies of multi-finger synergies during isometric pressing tasks. In those studies, synergies were defined based on the principle of motor abundance (Gelfand and Latash 1998; Latash 2012) as neural organizations that ensure stability of a performance variable produced by a redundant set of elements. Multi-finger synergies were quantified using the computational apparatus associated with the uncontrolled manifold (UCM) hypothesis (Scholz and Schöner 1999). The synergy index showed a drop about 200–250 ms prior to the initiation of a quick force pulse from a steady-state force level when the subjects were free to initiate the pulse at any time.

In those first studies, a hypothesis was offered that the purpose of ASAs was to attenuate pre-existent synergies stabilizing a steady-state value of a task-related performance variable in preparation to a quick change in that variable. Further, similarities between ASAs and anticipatory postural adjustments (APAs, reviewed in Massion 1992) have been noted. In particular, both ASAs and APAs are delayed when the same action is performed under typical simple reaction time conditions (Lee et al. 1987; DeWolf et al. 1998; Olafsdottir et al. 2005), both ASAs and APAs are delayed and reduced in magnitude in healthy older adults (Woollacott et al. 1988; Olafsdottir et al. 2007), and both are reduced in patients with Parkinson’s disease (Bazalgette et al. 1986; Park et al. 2012). These observations led to a hypothesis that the two apparently different reflections of anticipatory motor control processes could result from a single neural process with two distinct phases. Indeed, several recent studies of multi-muscle synergies in standing subjects showed a consistent sequence of events with postural adjustments preceded by ASAs (Klous et al. 2011; Krishnan et al. 2011, 2012).

To our knowledge, the current study is the first to show that ASAs can be seen in conditions when APAs should not be expected. Indeed, if action direction is unknown, it is counterproductive to generate net forces in preparation to the action. However, it is not counterproductive to facilitate future action by attenuating in advance synergies stabilizing the relevant performance variable. Otherwise, the subject would be fighting his/her own synergies. So, under conditions when the timing of an action is known while its direction is not, ASAs become the only anticipatory mechanism that makes sense. The results of our current study show that ASAs can indeed be seen with only slightly changed characteristics in such conditions.

These findings fit well the central back-coupling hypothesis on the origin of multi-element synergies (Latash et al. 2005; cf. the idea of planning in the space of elements, Desmurget et al. 1995) mentioned in the Introduction. According to this hypothesis, the hierarchically higher neural controller manipulates two groups of neural variables, NV1 and NV2. NV1 define magnitudes of task-specific salient performance variables. Changes in NV1 in anticipation of an action or perturbation lead to APAs. NV2 define stability properties of those same performance variables; this is done through adjustments of the gains in the hypothesized loops that provide the back-coupling within the hypothesis. An early change in NV2 results in ASAs.

In most everyday actions, if a person plans a quick action, a sequence of events happens. First, NV2 change and thus facilitate a quick change in the relevant variable. Then, the variable itself is changed (NV1 change), which may be reflected in APAs. In this scheme, changes in NV2 and NV1 are time linked. In some situations, however, one knows that a quick action will be necessary at a certain time but does not know exactly the direction of the action. For example, if one prepares to take a tennis serve, the timing of the serve may be well-known to the player watching the opponent, while it is sometimes impossible to predict whether the serve will require a forehand or a backhand response. One useful strategy in such a situation would be to attenuate synergies stabilizing the whole-body and the arm posture. Turning the synergies off completely is not an option since then losing balance becomes quite probable. Attenuating those synergies is expected to lead to higher body-arm sway, which is quite frequently observed in tennis players who are getting ready for a high-speed serve. A similar increase in the body sway can be observed in football (soccer) goalkeepers getting ready for a penalty shot.

Under conditions when the subjects did not know in advance the exact timing of the “perturbation”, we still observed changes in the synergy index (ASAs) prior to the action initiation. Under those conditions, ASAs were delayed and much smaller in magnitude but, nevertheless, they were present. There are two possible, mutually non-exclusive, reasons for these observations. First, the subjects knew that a “perturbation” would take place at some time within a certain time window. So, they could be preparing for such “perturbations” by slowly decreasing the synergy index over the waiting time. These adjustments could become stronger towards the end of the trial time window because of the increase in the probability of a perturbation. Such a strategy is compatible with observations on changes in the synergy index in conditions when the target of a quick movement could jump (De Freitas et al. 2007; De Freitas and Scholz 2009). Probably, such slow drifts (that were beyond our current means of detection) differed across subjects. As a result, statistical analysis showed that they led to a significant drop in the synergy index only very close to the action initiation. Producing gain changes only in a fraction of the trials could potentially eliminate ASAs; unfortunately, this would have increased the total number of trials and might lead to fatigue.

The second possibility is related to the fact that the force adjustment time values in our study were much longer than in the earlier studies of Olafsdottir and her colleagues (2005, 2007). While under simple reaction time conditions, ASAs were not observed (Olafsdottir et al. 2005, 2007), a two-choice condition could have given subjects enough time to start initiating ASAs (cf. Slijper et al. 2002).

Effects of incomplete advance knowledge on early and late reaction to perturbations have been studied by Grierson et al. (2009, 2011; also see Grierson and Elliott 2008). Those studies explored early and late corrective actions to mechanical perturbations and changes in target location. The subjects could show early movement adjustments based on the mere recognition that expectations have been violated. While there is a common feature between those studies and the current paper, namely the emphasis on the role of prior information on motor adjustments, the focus of our study differs from that of the mentioned papers: Our study is primarily focused on the drop in the synergy index prior to the visual gain change (ASA), not on reactions to the gain change.

While there were differences between the actual total force (F_TOT) and feedback force (F_FB) following a change in k, both F_TOT and F_FB represented weighted sums of the individual finger forces. As a result, negative co-variation among finger forces across trials would be expected to contribute to higher synergy indices, V with respect to both those variables. Our analysis of ΔV 1 s after the “perturbation” showed that the subjects co-varied commands to individual fingers in a more subtle way. Indeed, V computed with respect to F_FB was consistently higher as compared to ΔV computed with respect to F_TOT. This result is far from being trivial. Note that the proprioceptive feedback from the fingertips was unchanged under all conditions. If force-stabilizing synergies were built on feedback loops from proprioceptors, one could have expected higher ΔV values for F_TOT, which represents the sum of actual finger forces. The opposite result suggests that such synergies are built on essentially feed-forward mechanisms (cf. Goodman and Latash 2006) that may be using back-coupling loops within the central nervous system. Schemes built on this principle have been suggested compatible with the idea of motor control with referent body configurations (Martin et al. 2009; Latash 2010; cf. Feldman et al. 2007; Feldman 2011).

According to the referent configuration hypothesis, neural control of a natural movement is based on specifying neural variables that lead to changes in a set of referent values for a salient, task-specific variable, given the external force field. Further, few-to-many mappings result in referent values for elemental variables that contribute to that salient variable. Those mappings may be organized in a synergic way, that is, they allow relatively large variations in the elemental variables as long as the referent value for the salient variable remains relatively unchanged. Adjustments in such mappings without a change in the referent value for the salient variable are plausible origins of ASAs.

To conclude, our study has shown that ASAs can take place in conditions when action direction is unknown. These results show that ASAs are an important mechanism of anticipatory motor control that is likely to be used under a variety of conditions when a quick action by a redundant set of elements is prepared. (NB: All natural human actions are produced by redundant sets of elements; Bernstein 1967).