Memory is strongly affected by the context in which a particular memory is formed and remembered. For example, visiting a familiar place can often trigger memories associated or “tagged” with that place. Such tagging also exists for memories related to movement: for instance, distinct motor memories for a limb movement are formed depending on whether the other limb is stationary or moving. However, little is known about how the tagging of such motor memories takes place.

Nozaki et al. have now used a technique known as transcranial direct current stimulation to generate artificial “tags” for motor memories. In the experiments, volunteers tried to move a robotic arm towards a goal while the robot pushed their hand off-course. Sometimes the robot pushed the participant’s hand to the left, and sometimes to the right. This makes the task difficult to learn, even when the cue for the direction is provided, as the motor memories that are made to counteract each push overwrite each other.

Nozaki et al. used transcranial stimulation to alter the background electrical activity in the sensorimotor regions of the participants’ brains as they performed the robotic arm task. Artificially generating a different pattern of background brain electrical activity for each push direction caused the motor memories associated with leftward and rightward pushes to be tagged differently. Once this association had been learnt, applying the artificial brain stimulation pattern associated with one of the pushes resulted in the participants unconsciously compensating for a push in that direction, even when it was not there.

Overall, the results presented by Nozaki et al. suggest that the background electrical activity seen in the brain can influence how a motor memory is created and later recalled. A future challenge is to investigate whether this technique could be used to help athletes improve their performance or to treat people with movement disorders. Further experiments are also needed to test whether the same approach can influence the formation and recollection of other kinds of memories, such as those related to fear.

Context influences memory encoding and retrieval, including fear conditioning responses in rodents (Maren et al., 2013) and declarative memory in humans (Godden and Baddeley, 1975). Although previous studies have examined whether context-dependency also exists in motor memory, context-dependent motor learning based on contextual cues has proven quite difficult. For example, participants have difficulty adapting identical reaching movements to conflicting dynamical environments (e.g., velocity-dependent rightward and leftward force fields) according to a contextual cue (e.g., room color, target color, etc [Gandolfo et al., 1996; Osu et al., 2004; Nozaki et al., 2006; Hirashima and Nozaki, 2012; Kadota et al., 2014]).

It is only recently that a wide variety of contexts helpful to create distinct motor memories has been discovered (Osu et al., 2004; Nozaki et al., 2006; Hirashima and Nozaki, 2012; Kadota et al., 2014; Cothros et al., 2009; Ikegami et al., 2010; Howard et al., 2011; Yokoi et al., 2011; Howard et al., 2012, 2010; Sarwary et al., 2013; Yokoi et al., 2014; Howard et al., 2015; Sarwary et al., 2015). However, the underlying mechanisms regarding context-dependent motor learning and memory are largely unknown. Notably, contexts that are shown to be useful are often associated with different neural activity patterns of the sensorimotor cortex including the primary motor cortex (M1) and the premotor cortex (PM). For example, distinct motor memories for identical reaching movements can be created depending on whether the opposite arm is stationary or moving (Nozaki et al., 2006; Kadota et al., 2014; Nozaki and Scott, 2009). Consistent with this finding, it has been reported that opposite arm movements alter M1 and PM activity during reaching movements (Donchin et al., 2001, 2002; Cisek et al., 2003; Ganguly et al., 2009; Rokni et al., 2003). In agreement with the significant role of these brain areas in motor learning (Kadota et al., 2014; Gandolfo et al., 2000; Li et al., 2001; Muellbacher et al., 2002; Arce et al., 2010; Orban de Xivry et al., 2011; 2011; 2013), we hypothesized that, when a motor memory is formed under different activity patterns in the sensorimotor cortex associated with different contexts, a motor memory specific to this particular activity pattern is created. As a result, later reinstating this activity pattern in these areas should lead to automatic retrieval of the corresponding memory.

Here, we tested this hypothesis directly using transcranial direct current stimulation (tDCS) (Nitsche et al., 2008; Orban de Xivry and Shadmehr, 2014; Di Lazzaro and Rothwell, 2014). tDCS to the sensorimotor cortex modulates spontaneous M1 activity and excitability according to its polarity. As a result, the size of the motor evoked potential induced by transcranial magnetic stimulation is increased or decreased by, respectively, anodal or cathodal tDCS during (Nitsche et al., 2005; 2007) and even after the application of stimulation (Nitsche and Paulus, 2000; 2001; Siebner et al., 2004). This suggests that different brain activity patterns (i.e. specific to the polarity of the stimulation) could be artificially created by tDCS. We predicted that motor learning performed under different background activity patterns of the sensorimotor cortex created by tDCS would yield to the formation of separate motor memories. Participants were trained to perform reaching movements in the presence of two conflicting force fields while receiving tDCS on the sensorimotor cortex. Critically, training for each force field was always associated with a distinct tDCS polarity. We then examined if, after training, applying a particular tDCS polarity reactivated the motor memory associated with this polarity, despite no explicit contextual change.

Participants performed reaching movements with their right arm in the presence of two conflicting force fields while receiving anodal or cathodal tDCS to the left M1 associated with each force field (training period: 12 blocks of 22 trials). We stimulated M1 because it is known to play a significant role in motor learning (Kadota et al., 2014; Gandolfo et al., 2000; Li et al., 2001; Muellbacher et al., 2002; Arce et al., 2010; Orban de Xivry et al., 2011, 2011, 2013). One training block consisted of 18 reaching movements toward a target (movement distance was 10 cm), in either a rightward or leftward velocity-dependent force field (Shadmehr and Mussa-Ivaldi, 1994), and four error-clamp trials (first and last two trials of the block, Figure 1a,c) to avoid unnecessary learning that could occur during the switch of tDCS polarity. Critically, this polarity (2 mA, bi-hemispheric montage with electrodes on the left and right M1) changed in congruence with the direction of the force field (Figure 1a,c) in order to tag motor memories with stimulation polarity. In other words, in this first experiment, anodal and cathodal stimulation of the left M1 were always associated with rightward and leftward force field perturbations, respectively.

Figure 1

Download asset Open asset

During a subsequent test period, we investigated whether modulating activity patterns in the sensorimotor cortex by anodal or cathodal tDCS could induce artificial recollection of the corresponding motor memory. During that period, tDCS polarity alternated between anodal and cathodal, as it did during the training period, but there was no force field perturbation (4 blocks; Figure 1b,d). Instead, a series of error-clamp trials was used to evaluate motor memory content throughout the test period (Figure 1b,d). If, during training, motor memories were tagged according to tDCS polarity, we should observe leftward force output (this direction was defined as negative) to compensate for the rightward force field during anodal tDCS. On the other hand, application of cathodal tDCS was expected to induce force output in a positive (i.e., rightward) direction. It should be noted that no cognitive and/or contextual cues were provided throughout the experiments.

tDCS induces motor memory retrieval during the test period

During the test period, the first eight participants received anodal and cathodal tDCS in alternation, starting with anodal stimulation (Figure 1g, T-T_ACAC). In this group, evolution of the force output during the test period (Figure 2a, dashed line and open circles) indicates the presence of a tDCS effect. As predicted, the force output exhibited a clear polarity-dependent change, and the forces during cathodal stimulation (blue open circles in Figure 2a) were larger than the forces during anodal stimulation (red open circles in Figure 2a). Another group of participants (N = 8, T-T_CACA, Figure 1g) were trained with an identical protocol but first received cathodal stimulation during the test period. This subgroup also exhibited clear polarity-dependent effects on force output during the test period (Figure 2a, solid line and filled circles), which was consistent with our predictions. In addition, these changes were in anti-phase for the two groups.

Figure 2 with 2 supplements see all

Download asset Open asset

Natural exponential decay in force output observed at the beginning of the test period co-occurs with the polarity-dependent effect of tDCS. We took it into account with two different methods. In the first model-free approach, we tried to eliminate this decay effect by contrasting the force output of the two subgroups (Figure 2 ΔForce obtained by subtracting the data of the ACAC subgroup from the data of the CACA subgroup). This method relies on the assumption that exponential decay was similar in both subgroups. In the second model-based approach, a summation of two exponential curves was fitted to the force output data of each participant during the test period and we analyzed the residuals around the curve in function of tDCS polarity (Figure 3). This method bears the advantage of being applied on each subgroup separately.

Figure 3 with 3 supplements see all

Download asset Open asset

Figure 2b represents the trial-dependent change in the ΔForce calculated using a bootstrap method. If the stimulation had no effect on motor memories, the evolution of ΔForce over the test period should be flat. However, we observed a clear modulation pattern (Figure 2b,c): the ΔForce was smaller for the first and third block compared to the second and fourth block. This effect of block order (Figure 2—figure supplement 1) on the ΔForce was statistically significant (Figure 2c; permutation test: p = 0.0009). In addition, this effect appeared stable from the first to the second half of the test period (permutation test: interaction between period and block order: p = 0.313).

Importantly, these changes were consistent with the pattern predicted by the training. For example, in the first block, the force output of T-T_ACAC group should be more leftward than that of T-T_CACA group and thus the ΔForce was more negative. The degree of ΔForce modulation from the first to second block, from the second to third block and from third to fourth block (the value was defined positive if the change was congruent with the predicted change) was 0.57 ± 0.32 (permutation test: p = 0.0632), 0.73 ± 0.21 (p = 0.0044), and 0.78 ± 0.13 N (p = 0.0001), respectively (mean ± standard deviation of the bootstrapped samples). In our experimental setting, the force output of full adaptation varied from -4 N to 4 N (viscosity was 10 N/(m/s) and the peak velocity of the handle was 0.4 m/s). Thus, approximately 7–10% force modulation was induced via tDCS for the T-T group.

Given the effect of tDCS for muscular force output (Salimpour and Shadmehr, 2014; Tanaka et al., 2009), one might expect that the movement kinematics were also modulated by the polarity of the stimulation. Such modulation might confound the actual effect of tDCS on force output because the perturbation was a velocity-dependent force-field. However, we did not observe any polarity-dependent changes in movement peak velocity [Figure 2—figure supplement 2; interaction between group and block order by repeated measures 3-way ANOVA: F(1,14) = 0.374, p = 0.551, ${\displaystyle {\eta }_p^2}$ = 0.024; main effect of block order: F(1,14) = 3.076, p = 0.101, ${\displaystyle {\eta }_p^2}$ = 0.18]. This demonstrates that polarity-dependent changes in force output were not caused by hand kinematic modulation due to tDCS.

We also characterized the effect of tDCS by examining variation in force output around the natural exponential decay of motor memories during the test period (model-based approach). If the stimulation had no effect on motor memories, the decay of the force output should simply follow this exponential function and the residuals should be evenly distributed around the exponential curve in all blocks (Figure 3—figure supplement 1a). In contrast, if the stimulation polarity influences the force output during the test period, the residuals should be more positive (above the exponential curve) during cathodal stimulation and more negative (below the exponential curve) during anodal stimulation (Figure 3—figure supplement 1b). Given that the two subgroups received different polarity during each of the blocks, the polarity-dependent change in residuals was opposite between the two groups (Figure 3a). That is, the residuals were always more positive for the group receiving cathodal stimulation during one of the blocks and more negative for the group receiving anodal stimulation during the same block (Figure 3a). The modulation of the force output by tDCS was analyzed with a 3-way ANOVA with subgroup as a between-subject factor and period (first or second half of the test period, 2 blocks each) and block-order (first and second block of each half of the test period) as a within-subject factors (See Figure 2—figure supplement 1 for the definitions of period and block order). Given that the polarity was opposite across groups in function of the block order factor (anodal in the first and third block for T-T_ACAC but in the second and fourth block for T-T_CACA), the modulation of force output by tDCS resulted in a significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F(1,14) = 39.24, p = 2.08 × 10^–5, ${\displaystyle {\eta }_p^2}$ = 0.74]. For each subgroup separately, there was a polarity-dependent change in the residuals as indicated by the main effect of block order on the residuals [T-T_ACAC: F(1,14) = 12.70, p = 3.10 × 10^–3, ${\displaystyle {\eta }_p^2}$ = 0.47; T-T_CACA: F(1,14) = 28.04, p = 1.13 × 10^–4, ${\displaystyle {\eta }_p^2}$ = 0.67]. This effect indicates that, for each subgroup, the force output varied with the polarity of the stimulation.

The modulation of force output with tDCS polarity was also examined when the association between tDCS polarity and force direction was reversed (T^ffrev-T group), i.e., when anodal and cathodal stimulations were associated with leftward and rightward force fields, respectively (Figure 3—figure supplement 2). While the effect appeared clear in one of the subgroups (T^ffrev-T_CACA), it was absent in the other one (T^ffrev-T_ACAC). This mixed effect yielded a non-significant effect of block order on the ΔForce because the permutation test was based on the two groups, but there was still a significant change in ΔForce from the second to third block (Figure 3—figure supplement 2b,c; permutation test: p = 0.0127).

The polarity dependent modulation was more apparent in the model-based approach because it allowed us to study each group separately. In this case, we found a significant interaction between subgroup and block-order [repeated measures 3-way ANOVA: F(1,16) = 5.844, p = 0.028, ${\displaystyle {\eta }_p^2}$ = 0.27]. This interaction stems from the polarity-dependent changes in force output observed in the T^ffrev-T_CACA subgroup [main effect of block order on the residuals: F(1,16) = 7.58, p = 0.014, ${\displaystyle {\eta }_p^2}$ = 0.32]. This effect was identical to the one observed for the two T-T subgroups (Figure 3a). In contrast, for the other subgroup (T^ffrev-T_ACAC), there was no clear effect of stimulation polarity during the test period [simple main effect of block-order on the residuals: F(1,16) = 0.44, p = 0.515, ${\displaystyle {\eta }_p^2}$ = 0.027].

Overall, we recorded data from four groups in the active tDCS conditions (i.e., T-T and T^ffrev-T groups) and our ANOVA results revealed a significant effect in three out of the four groups. Given our statistical power and effect size, this is exactly what would be expected. We compute the probability that there was an effect of stimulation polarity on force during the test period when the effect is observed in 3 out of the 4 groups. To compute this, we used a rationale provided by Ioannidis (2005) but adapted this justification to our positive results, as was done by Lakens and Evers (2014). Given a power of 0.8, the probability of observing three significant and one non-significant finding if there is a true effect is as follows (Type 2 error): 0.8 × 0.8 × 0.8 × 0.2 = 0.1024. Any of the four groups could yield a non-significant finding, so the a-priori likelihood of finding three out of four significant effects is 0.4096. We also needed to find the probability of observing these results if there was no effect (null hypothesis is true). Given a Type 1 error of 0.05 (significance threshold), this probability is as follows: p = 0.05 × 0.05 × 0.05 × 0.95 = 0.00011875. Again, because any group could be non-significant, the probability of finding these results if there is no effect is p = 0.000475. With these data, we can compute the positive predictive value (PPV) (Ioannidis, 2005). This number represents the post-study probability that the effect is true: PPV = 0.4096/(0.4096 + 0.000475) = 0.998. That is, the likelihood that there is a true effect despite the fact that one of the four groups yielded a non-significant effect is 99.8%.

Active tDCS during the training and test periods is required for tagging and retrieval of motor memories

Polarity-dependent modulation in force output could reflect the effect of anodal and cathodal tDCS in facilitating or suppressing force output. However, this is unlikely, as subgroups in which sham tDCS (tDCS applied only at the beginning part of each block) was applied during the training period followed by active stimulation during the test period (S-T group: S-T_ACAC and S-T_CACA subgroups, N = 8 for each subgroup, Figure 1e,h) showed little influence of tDCS polarity on force output during the test period (Figure 2d). While we detected a main effect of block order on the ΔForce permutation test: p = 0.0263), this effect was reduced over time (permutation test: interaction between period and block order: p = 0.025) (Figure 2e,f). Indeed, ΔForce did not exhibit significant block-dependent modulation from the second to third block (permutation test: p = 0.905) and from the third to fourth block (p = 0.278), although this modulation was significant from the first to second block (p = 0.0132) (Figure 2f). We also compared the strength of modulation by subtracting the ΔForce of S-T group from ΔForce of T-T group (Figure 4a). Although we did not find the expected interaction between group and block order (p = 0.205), this appeared to change over time (permutation test: interaction between block order, period and group: p = 0.0595). Indeed, the modulation of ΔForce from the second to third block and from the third to fourth block was significantly larger in the T-T group than in the S-T group (Figure 4b; permutation test: p = 0.0012 and p = 0.0146, respectively). Together, this data revealed that the modulation of the force output of tDCS was larger in the T-T group than in the S-T group from the second block onwards.

Figure 4

Download asset Open asset

The significant change in ΔForce from the first to second block in the S-T group (Figure 2e,f) and the absence of the differences between T-T and S-T groups for these blocks (Figure 4a,b; permutation test: p = 0.36) indicate that the application of tDCS immediately after training can influence the exponential decay of motor memory in a manner opposite to the conventional effect of tDCS. Namely, cathodal stimulation makes motor memories last longer than anodal tDCS. Alternatively, the observed effect could reflect the possibility that the motor memory was tagged with the brain activity without tDCS and the application of different tDCS polarities during the test period might differently recruit the motor memories. Either way, these effects were not strong enough to induce block-dependent modulation in the later blocks (Figure 2e,f; Figure 4a,b).

These results were confirmed by the model-based approach. The residuals around the exponential curves in the S-T group did not exhibit any evidence of polarity-dependent modulation of force output [repeated measures 3-way ANOVA: interaction between subgroup and block order: F(1,14) = 0.005, p = 0.945, ${\displaystyle {\eta }_p^2}$ = 0.063] (Figure 3b). This discrepancy with the model-free approach is related to the ability of the exponential function to capture the early difference in decay rates. The modulation of the residuals via stimulation polarity was also larger for the T-T group than for the S-T group [interaction between group, subgroup, and block order by repeated measures 4-way ANOVA: F(1,28) = 16.80, p = 2.42 × 10^–4, ${\displaystyle {\eta }_p^2}$ = 0.37]. These results confirm that the tDCS effect was restricted to the group where active stimulation was received during the training and test periods.

In another control experiment, active stimulation was applied during the training period while sham tDCS was applied during the test period (T-S group) (Figure 1f,i). In this group, we did not observe any polarity-dependent modulation of force output with tDCS polarity (Figure 2g). Indeed, the model-free approach applied to ΔForce indicated that there was no significant block-dependent modulation (Figure 2h,i; permutation test: main effect of block order: p = 0.163). The modulation of ΔForce was significantly larger in the T-T group than in the T-S group (Figure 4c,d; permutation test, interaction between group and block order: p = 0.0018). In addition, the modulation of ΔForce was significant from the first to second block, from the second to third block, and from the third to fourth block in the T-T group than in the S-T group (Figure 4d; permutation test: p = 0.0573, 0.0213, and 0.0014, respectively).

These results were confirmed by the model-based approach. For the T-S group, the residuals around the exponential curve were not modulated by tDCS polarity [Figure 3c, repeated measures 3-way ANOVA, interaction between subgroup and block order: F(1,14) = 2.33, p = 0.149, ${\displaystyle {\eta }_p^2}$ = 0.14). In addition, the polarity-dependent modulation of force stimulation observed in the T-T group was significantly larger than the modulation observed in the T-S group [interaction between group, subgroup, and block order by repeated measures 4-way ANOVA: F(1,28) = 29.79, p = 7.93 × 10^–6, ${\displaystyle {\eta }_p^2}$ = 0.52]. Together, these results indicate that activity patterns in the sensorimotor cortex during motor learning needs to be reinstated in order to retrieve those memories.

PPC tDCS is not effective to tag and retrieve motor memories

Finally, to assess the influence of the electrode position on the polarity-dependent modulation of force output, we performed an additional control experiment in which tDCS was applied to the posterior parietal cortex (PPC: PPC group) (Figure 5a). Stimulation of the PPC was unable to yield block-dependent modulation of ΔForce as revealed by the model-free approach (Figure 5b,c; permutation test: p = 0.369). In addition, block-dependent changes of ΔForce were significantly smaller in the PPC group than in the T-T group (Figure 5e,f; permutation test, interaction between group and block order: p = 0.0082), although the changes of force output from the first to second (permutation test: p = 0.0538) from the second to third (p = 0.113) and from the third to fourth block (p = 0.153) did not reach the significant level. Similarly, the model-based approach confirmed the absence of effect of tDCS when applied on PPC [Figure 5d, repeated measures 3-way ANOVA on the residuals, interaction between subgroup and block order: F(1,16) = 1.98, p = 0.179, ${\displaystyle {\eta }_p^2}$ = 0.11]. In addition, repeated measure 4-way ANOVA indicated that polarity-dependent modulation of force output was significantly greater for the T-T group than for the PPC group [interaction between group, subgroup, and block order: F(1,30) = 5.10, p = 0.031, ${\displaystyle {\eta }_p^2}$ = 0.15], indicating that M1 stimulation was more effective than PPC stimulation in creating and retrieving polarity-dependent motor memories.

Figure 5

Download asset Open asset

tDCS effects during the training period

We also examined the potential effect of tDCS polarity alternations on the interference between opposing perturbations during the training period (Figure 6). To investigate the amount of interference between the two perturbations, we measured the lateral deviations of the first trial during rightward and leftward force-field training. With this measure, higher interference is associated with higher lateral deviation on the first trial. When comparing this measure between the T-T, S-T, and T-S groups, we did not detect any significant differences in lateral deviation during the first trial [repeated measures 3-way ANOVA: F(2,45) = 2.369, p = 0.105, ${\displaystyle {\eta }_p^2}$ = 0.088] (Figure 6b). However, when the data from the T-T and T-S groups were grouped together (they received the same training and stimulation during the training period), lateral deviations during the first trial were significantly smaller (i.e., less interference) for groups who received active tDCS whilst training (T-T and T-S groups) than for the sham group (S-T group) [main effect of group: F(1,46) = 4.435 p = 0.041, ${\displaystyle {\eta }_p^2}$ = 0.10; interaction between group and force-field direction (or tDCS polarity): F(1,46) = 0.148, p = 0.708, ${\displaystyle {\eta }_p^2}$ = 0.003]. This suggests that tagging motor memories with different tDCS might have an effect during the training period. However, this result needs to be confirmed in follow-up studies given the limited effect size and borderline significance here. In contrast, lateral deviation at the end of each block was not different between groups [repeated measures 3-way ANOVA, main effect of group: F(1,46) = 0.574, p = 0.453, ${\displaystyle {\eta }_p^2}$ = 0.012; interaction between group and force-field direction: F(1,46) = 0.401, p = 0.53, ${\displaystyle {\eta }_p^2}$ = 0.009] (Figure 6c).

Figure 6

Download asset Open asset

Recent studies have demonstrated several striking examples of context-dependent motor learning (Osu et al., 2004; Nozaki et al., 2006; Hirashima and Nozaki, 2012; Kadota et al., 2014; Cothros et al., 2009; Ikegami et al., 2010; Howard et al., 2011; Yokoi et al., 2011; Howard et al., 2012; Yokoi et al., 2014; Howard et al., 2015). Distinct motor memories for identical movements can be formed and retrieved according to whether the opposite arm is stationary (i.e., unimanual) or moving (i.e., bimanual) (Nozaki et al., 2006; Kadota et al., 2014; Nozaki and Scott, 2009). Motor memories can also be influenced by movement direction of the opposite arm (Yokoi et al., 2011; Howard et al., 2010; Yokoi et al., 2014), whether the movement is discrete or rhythmic (Ikegami et al., 2010; Howard et al., 2011), or how follow-through movements are performed (Howard et al., 2015). Although this type of learning enables us to perform flexible actions for adapting to a wide variety of dynamical environments (Yokoi et al., 2011; 2014), it remains unknown how different behaviors result in the formation of distinct motor memories.

Intriguingly, previous neurophysiological and/or brain imaging studies have shown that activity in motor-related neural areas also differs according to how an opposite arm is moving (Donchin et al., 2001, 2002; Cisek et al., 2003; Ganguly et al., 2009), between discrete and rhythmic movement (Schaal et al., 2004), or the direction of the follow-through movement (Baldauf et al., 2008). It should be noted that previous studies have also suggested that movement adaptation to a novel dynamical environment is accomplished through neural activity changes in these motor-related areas (Kadota et al., 2014; Gandolfo et al., 2000; Li et al., 2001; Arce et al., 2010). Thus, it is natural to speculate that these neurons change their activation patterns during motor adaptation, depending on how they are originally recruited within the behavioral context. For example, distinct motor memories for unimanual and bimanual movements (Nozaki et al., 2006) can be explained by considering that partially different populations of M1 and PM neurons are involved for motor adaptation (Nozaki and Scott, 2009).

We hypothesized that motor learning performed during distinct background neural activity in the sensorimotor cortex leads to the development of a motor memory that is specifically associated with a particular behavior. We examined this hypothesis by artificially creating two different background activity patterns in the sensorimotor cortex, using anodal and cathodal tDCS, when participants learned to perform reaching movements within 2 conflicting forcefields. Most importantly, each tDCS polarity was always associated with one of two force-fields. Consistent with this hypothesis, our experiments demonstrated that, after training, force output during the test period was modulated by the corresponding polarity (T-T group; Figure 2a–c, Figure 3a). In contrast, force output modulation was not observed when tDCS was applied to the PPC (PPC group; Figure 5). Furthermore, the ability of tDCS to modulate force output, depending on polarity through mere changes in cortical excitability (Salimpour and Shadmehr, 2014; Tanaka et al., 2009), was insufficient to explain our observed force output modulation (S-T group; Figure 2d–f, Figure 4a,b).

Participants were also unlikely to modulate the force output according to tDCS-induced skin sensation, as both the PPC and T-T groups experienced similar tDCS sensations throughout the experiments. Taken together, our experiments strongly indicate that differential background activity patterns in the sensorimotor cortex, artificially created when a motor learning task is performed, could lead to the formation of distinct motor memories. These memories could then be artificially retrieved by applying the same-polarity tDCS during a reaching movement.

It is important to emphasize that our experimental paradigm fully departs from conventional tDCS applications, which are traditionally restricted to increasing/decreasing motor output and accuracy, motor memory consolidation, rehabilitation outcomes, and memory functioning (Nitsche et al., 2008; Orban de Xivry and Shadmehr, 2014). Furthermore, if tDCS can artificially induce motor memory recollection, tDCS may also artificially maintain motor memories when tested in different contexts. Indeed, we have previously shown that reaching movement adaptation to a force-field, while receiving anodal or cathodal tDCS, is more strongly generalized to reaching movements toward another direction performed with different arm postures (Orban de Xivry et al., 2011). This improved generalization could be explained by the artificial maintenance of activity patterns in the sensorimotor cortex with tDCS when transitioning from the adaptation phase to the generalization phase.

We assumed that tDCS only influenced neural activity in the sensorimotor cortex, particularly in M1 and PM neurons. Recent studies using fMRI (Lindenberg et al., 2013) and electrical fields models (Gillick et al., 2014) have indicated that bihemispheric tDCS effects are more localized than when using a conventional stimulation montage (i.e., placing one electrode on the forehead), but we cannot reject the possibility that broader brain areas are also involved. Given our bilateral stimulation, it is also possible that stimulation to the ipsilateral right sensorimotor cortex (ipsilateral to the moving arm) caused the observed effects (rather than the contralateral left sensorimotor cortex). Further studies using a High-Defintion tDCS (Datta et al., 2009) will be necessary to clarify these issues. Finally, however, given that tDCS had no effect on memory retrieval when applied on the PPC, the involvement of the underlying parietal areas, including sensory areas, is unlikely or relatively small.

We also assumed that we could create different background activity patterns via anodal and cathodal tDCS. Indeed, previous studies have demonstrated that M1 excitability (Nitsche et al., 2005; Nitsche et al., 2007) or neuronal firing rates (Creutzfeldt et al., 1962) could be modulated according to tDCS polarities. Although the present study only focused on the immediate modulating effect of tDCS on neural activity, tDCS also has long-lasting or plastic effects on the cortex (Nitsche et al., 2008), raising questions as to whether activity patterns in the sensorimotor cortex were similar during the training and testing periods. For example, M1 excitability remains elevated, even after anodal tDCS termination, when stimulus duration is sufficiently long (Nitsche et al., 2007; 2008; Nitsche and Paulus, 2000; 2001). However, we considered that any long-lasting effects did not play a dominant role in our results. First, as tDCS polarity switched every 2 min, durations were too short to induce long-lasting effects on cortical excitability. Second, as anodal and cathodal tDCS alternated, any long-lasting effects, including homeostatic plasticity (Siebner et al., 2004), were likely cancelled out.

Previous studies have suggested that reinstating the same activity pattern as was present during learning might contribute to one’s ability to recall that memory (Maren et al., 2013; Fanselow, 2010). For example, invasive electrical stimulation to a particular region of the temporal cortex among patients with epilepsy can induce declarative memory recollection. This is perhaps due to reproducing an activity pattern similar to what was observed during voluntary recollection of that same memory (Jacobs et al., 2012). Similarly, we used tDCS to reinstate an activity pattern in the sensorimotor cortex during learning in order to induce implicit retrieval of a particular motor memory. These effects are also similar to optogenetics’ ability to manipulate a fear memory in mice by directly altering hippocampal neural activity (Liu et al., 2012; Ramirez et al., 2013).

In summary, our study demonstrates a causal link between background neural activity in the sensorimotor cortex during learning and context-dependent motor memories. Here, we provided initial evidence that human motor memories can be artificially tagged and later retrieved via noninvasive brain stimulation. This manipulation sheds light on the possibility that manipulating the formation and recollection of various memories, including declarative and other types of motor memories (e.g., visuomotor rotation task or sequence learning), can be achieved by artificially changing activity patterns of the corresponding brain region during the learning period. Thus, our novel tDCS application opens up new avenues for implementing tDCS in human memory research.

The experiments were conducted in accordance with the Declaration of Helsinki. The ethics committee from The University of Tokyo approved all experimental procedures.

1. 1. Arce F
   2. Novick I
   3. Mandelblat-Cerf Y
   4. Vaadia E

   (2010) Neuronal correlates of memory formation in motor cortex after adaptation to force field

   Journal of Neuroscience 30:9189–9198.

   https://doi.org/10.1523/JNEUROSCI.1603-10.2010

   • Google Scholar
2. 1. Baldauf D
   2. Cui H
   3. Andersen RA

   (2008) The posterior parietal cortex encodes in parallel both goals for double-reach sequences

   Journal of Neuroscience 28:10081–10089.

   https://doi.org/10.1523/JNEUROSCI.3423-08.2008

   • Google Scholar
3. 1. Brennan AE
   2. Smith MA

   (2015) The decay of motor memories is independent of context change detection

   PLoS Computational Biology 11:e1004278.

   https://doi.org/10.1371/journal.pcbi.1004278

   • Google Scholar
4. 1. Cisek P
   2. Crammond DJ
   3. Kalaska JF

   (2003) Neural activity in primary motor and dorsal premotor cortex in reaching tasks with the contralateral versus ipsilateral arm

   Journal of Neurophysiology 89:922–942.

   https://doi.org/10.1152/jn.00607.2002

   • Google Scholar
5. 1. Cothros N
   2. Wong J
   3. Gribble PL

   (2009) Visual cues signaling object grasp reduce interference in motor learning

   Journal of Neurophysiology 102:2112–2120.

   https://doi.org/10.1152/jn.00493.2009

   • Google Scholar
6. 1. Creutzfeldt OD
   2. Fromm GH
   3. Kapp H

   (1962) Influence of transcortical d-c currents on cortical neuronal activity

   Experimental Neurology 5:436–452.

   https://doi.org/10.1016/0014-4886(62)90056-0

   • Google Scholar
7. 1. Criscimagna-Hemminger SE
   2. Shadmehr R

   (2008) Consolidation patterns of human motor memory

   Journal of Neuroscience 28:9610–9618.

   https://doi.org/10.1523/JNEUROSCI.3071-08.2008

   • Google Scholar
8. 1. Datta A
   2. Bansal V
   3. Diaz J
   4. Patel J
   5. Reato D
   6. Bikson M

   (2009) Gyri-precise head model of transcranial direct current stimulation: improved spatial focality using a ring electrode versus conventional rectangular pad

   Brain Stimulation 2:201–207.

   https://doi.org/10.1016/j.brs.2009.03.005

   • Google Scholar
9. 1. Di Lazzaro V
   2. Rothwell JC

   (2014) Corticospinal activity evoked and modulated by non-invasive stimulation of the intact human motor cortex

   Journal of Physiology 592:4115–4128.

   https://doi.org/10.1113/jphysiol.2014.274316

   • Google Scholar
10. 1. Donchin O
    2. Gribova A
    3. Steinberg O
    4. Bergman H
    5. Cardoso de Oliveira S
    6. Vaadia E

    (2001) Local field potentials related to bimanual movements in the primary and supplementary motor cortices

    Experimental Brain Research 140:46–55.

    https://doi.org/10.1007/s002210100784

    • Google Scholar
11. 1. Donchin O
    2. Gribova A
    3. Steinberg O
    4. Mitz AR
    5. Bergman H
    6. Vaadia E

    (2002) Single-unit activity related to bimanual arm movements in the primary and supplementary motor cortices

    Journal of Neurophysiology 88:3498–3517.

    https://doi.org/10.1152/jn.00335.2001

    • Google Scholar
12. 1. Fanselow MS

    (2010) From contextual fear to a dynamic view of memory systems

    Trends in Cognitive Sciences 14:7–15.

    https://doi.org/10.1016/j.tics.2009.10.008

    • Google Scholar
13. 1. Gandolfo F
    2. Li C
    3. Benda BJ
    4. Schioppa CP
    5. Bizzi E
    6. Li C-SR
    7. Li C-SR BBJ
    8. Padoa-Schioppa C

    (2000) Cortical correlates of learning in monkeys adapting to a new dynamical environment

    PNAS 97:2259–2263.

    https://doi.org/10.1073/pnas.040567097

    • Google Scholar
14. 1. Gandolfo F
    2. Mussa-Ivaldi FA
    3. Bizzi E

    (1996) Motor learning by field approximation

    PNAS 93:3843–3846.

    https://doi.org/10.1073/pnas.93.9.3843

    • Google Scholar
15. 1. Ganguly K
    2. Secundo L
    3. Ranade G
    4. Orsborn A
    5. Chang EF
    6. Dimitrov DF
    7. Wallis JD
    8. Barbaro NM
    9. Knight RT
    10. Carmena JM

    (2009) Cortical representation of ipsilateral arm movements in monkey and man

    Journal of Neuroscience 29:12948–12956.

    https://doi.org/10.1523/JNEUROSCI.2471-09.2009

    • Google Scholar
16. 1. Gillick BT
    2. Kirton A
    3. Carmel JB
    4. Minhas P
    5. Bikson M

    (2014) Pediatric stroke and transcranial direct current stimulation: methods for rational individualized dose optimization

    Frontiers in Human Neuroscience 8:.

    https://doi.org/10.3389/fnhum.2014.00739

    • Google Scholar
17. 1. Godden DR
    2. Baddeley AD

    (1975) Context-dependent memory in 2 natural environments - land and underwater

    British Journal of Psychology 66:325–331.

    https://doi.org/10.1111/j.2044-8295.1975.tb01468.x

    • Google Scholar
18. 1. Hesterberg T
    2. Monaghan S
    3. Moore DS
    4. Clipson A
    5. Epstein R

    (2003)

    The Practice of Business Statistics

    Bootstrap Methods and Permutation Tests, The Practice of Business Statistics, New York, W. H. Freeman and Company.

    • Google Scholar
19. 1. Hirashima M
    2. Nozaki D

    (2012) Distinct motor plans form and retrieve distinct motor memories for physically identical movements

    Current Biology 22:432–436.

    https://doi.org/10.1016/j.cub.2012.01.042

    • Google Scholar
20. 1. Howard IS
    2. Ingram JN
    3. Franklin DW
    4. Wolpert DM

    (2012) Gone in 0.6 seconds: the encoding of motor memories depends on recent sensorimotor states

    Journal of Neuroscience 32:12756–12768.

    https://doi.org/10.1523/JNEUROSCI.5909-11.2012

    • Google Scholar
21. 1. Howard IS
    2. Ingram JN
    3. Wolpert DM

    (2010) Context-dependent partitioning of motor learning in bimanual movements

    Journal of Neurophysiology 104:2082–2091.

    https://doi.org/10.1152/jn.00299.2010

    • Google Scholar
22. 1. Howard IS
    2. Ingram JN
    3. Wolpert DM

    (2011) Separate representations of dynamics in rhythmic and discrete movements: evidence from motor learning

    Journal of Neurophysiology 105:1722–1731.

    https://doi.org/10.1152/jn.00780.2010

    • Google Scholar
23. 1. Howard IS
    2. Wolpert DM
    3. Franklin DW

    (2015) The value of the follow-through derives from motor learning depending on future actions

    Current Biology 25:397–401.

    https://doi.org/10.1016/j.cub.2014.12.037

    • Google Scholar
24. 1. Ikegami T
    2. Hirashima M
    3. Taga G
    4. Nozaki D

    (2010) Asymmetric transfer of visuomotor learning between discrete and rhythmic movements

    Journal of Neuroscience 30:4515–4521.

    https://doi.org/10.1523/JNEUROSCI.3066-09.2010

    • Google Scholar
25. 1. Ioannidis JP

    (2005) Why most published research findings are false

    PLoS Medicine 2:e124.

    https://doi.org/10.1371/journal.pmed.0020124

    • Google Scholar
26. 1. Jacobs J
    2. Lega B
    3. Anderson C

    (2012) Explaining how brain stimulation can evoke memories

    Journal of Cognitive Neuroscience 24:553–563.

    https://doi.org/10.1162/jocn_a_00170

    • Google Scholar
27. 1. Kadota H
    2. Hirashima M
    3. Nozaki D

    (2014) Functional modulation of corticospinal excitability with adaptation of wrist movements to novel dynamical environments

    Journal of Neuroscience 34:12415–12424.

    https://doi.org/10.1523/JNEUROSCI.2565-13.2014

    • Google Scholar
28. Website

    1. Lakens D
    2. Evers ERK

    (2014) Too true to be bad:when sets of studies with significant and non-significant findings are probably true

    http://daniellakens.blogspot.co.uk/2014/06/too-true-to-be-badwhen-sets-of-studies.html
29. 1. Li C-SR
    2. Padoa-Schioppa C
    3. Bizzi E

    (2001) Neuronal correlates of motor performance and motor learning in the primary motor cortex of monkeys adapting to an external force field

    Neuron 30:593–607.

    https://doi.org/10.1016/S0896-6273(01)00301-4

    • Google Scholar
30. 1. Lindenberg R
    2. Nachtigall L
    3. Meinzer M
    4. Sieg MM
    5. Flöel A

    (2013) Differential effects of dual and unihemispheric motor cortex stimulation in older adults

    Journal of Neuroscience 33:9176–9183.

    https://doi.org/10.1523/JNEUROSCI.0055-13.2013

    • Google Scholar
31. 1. Liu X
    2. Ramirez S
    3. Pang PT
    4. Puryear CB
    5. Govindarajan A
    6. Deisseroth K
    7. Tonegawa S

    (2012) Optogenetic stimulation of a hippocampal engram activates fear memory recall

    Nature 484:381–385.

    https://doi.org/10.1038/nature11028

    • Google Scholar
32. 1. Maren S
    2. Phan KL
    3. Liberzon I

    (2013) The contextual brain: implications for fear conditioning, extinction and psychopathology

    Nature Reviews Neuroscience 14:417–428.

    https://doi.org/10.1038/nrn3492

    • Google Scholar
33. 1. Muellbacher W
    2. Ziemann U
    3. Wissel J
    4. Dang N
    5. Kofler M
    6. Facchini S
    7. Boroojerdi B
    8. Poewe W
    9. Hallett M

    (2002) Early consolidation in human primary motor cortex

    Nature 415:640–644.

    https://doi.org/10.1038/nature712

    • Google Scholar
34. 1. Nitsche MA
    2. Cohen LG
    3. Wassermann EM
    4. Priori A
    5. Lang N
    6. Antal A
    7. Paulus W
    8. Hummel F
    9. Boggio PS
    10. Fregni F
    11. Pascual-Leone A

    (2008) Transcranial direct current stimulation: State of the art 2008

    Brain Stimulation 1:206–223.

    https://doi.org/10.1016/j.brs.2008.06.004

    • Google Scholar
35. 1. Nitsche MA
    2. Doemkes S
    3. Karaköse T
    4. Antal A
    5. Liebetanz D
    6. Lang N
    7. Tergau F
    8. Paulus W

    (2007) Shaping the effects of transcranial direct current stimulation of the human motor cortex

    Journal of Neurophysiology 97:3109–3117.

    https://doi.org/10.1152/jn.01312.2006

    • Google Scholar
36. 1. Nitsche MA
    2. Paulus W

    (2000) Excitability changes induced in the human motor cortex by weak transcranial direct current stimulation

    Journal of Physiology 527 Pt 3:633–639.

    https://doi.org/10.1111/j.1469-7793.2000.t01-1-00633.x

    • Google Scholar
37. 1. Nitsche MA
    2. Paulus W

    (2001) Sustained excitability elevations induced by transcranial DC motor cortex stimulation in humans

    Neurology 57:1899–1901.

    https://doi.org/10.1212/WNL.57.10.1899

    • Google Scholar
38. 1. Nitsche MA
    2. Seeber A
    3. Frommann K
    4. Klein CC
    5. Rochford C
    6. Nitsche MS
    7. Fricke K
    8. Liebetanz D
    9. Lang N
    10. Antal A
    11. Paulus W
    12. Tergau F

    (2005) Modulating parameters of excitability during and after transcranial direct current stimulation of the human motor cortex

     Journal of Physiology 568:291–303.

    https://doi.org/10.1113/jphysiol.2005.092429

    • Google Scholar
39. 1. Nozaki D
    2. Kurtzer I
    3. Scott SH

    (2006) Limited transfer of learning between unimanual and bimanual skills within the same limb

    Nature Neuroscience 9:1364–1366.

    https://doi.org/10.1038/nn1785

    • Google Scholar
40. 1. Nozaki D
    2. Scott SH

    (2009) Multi-compartment model can explain partial transfer of learning within the same limb between unimanual and bimanual reaching

    Experimental Brain Research 194:451–463.

    https://doi.org/10.1007/s00221-009-1720-x

    • Google Scholar
41. 1. Orban de Xivry JJ
    2. Ahmadi-Pajouh MA
    3. Harran MD
    4. Salimpour Y
    5. Shadmehr R

    (2013) Changes in corticospinal excitability during reach adaptation in force fields

    Journal of Neurophysiology 109:124–136.

    https://doi.org/10.1152/jn.00785.2012

    • Google Scholar
42. 1. Orban de Xivry JJ
    2. Criscimagna-Hemminger SE
    3. Shadmehr R

    (2011) Contributions of the motor cortex to adaptive control of reaching depend on the perturbation schedule

    Cerebral Cortex 21:1475–1484.

    https://doi.org/10.1093/cercor/bhq192

    • Google Scholar
43. 1. Orban de Xivry JJ
    2. Marko MK
    3. Pekny SE
    4. Pastor D
    5. Izawa J
    6. Celnik P
    7. Shadmehr R

    (2011) Stimulation of the human motor cortex alters generalization patterns of motor learning

    Journal of Neuroscience 31:7102–7110.

    https://doi.org/10.1523/JNEUROSCI.0273-11.2011

    • Google Scholar
44. 1. Orban de Xivry JJ
    2. Shadmehr R

    (2014) Electrifying the motor engram: effects of tDCS on motor learning and control

    Experimental Brain Research 232:3379–3395.

    https://doi.org/10.1007/s00221-014-4087-6

    • Google Scholar
45. 1. Osu R
    2. Hirai S
    3. Yoshioka T
    4. Kawato M

    (2004) Random presentation enables subjects to adapt to two opposing forces on the hand

    Nature Neuroscience 7:111–112.

    https://doi.org/10.1038/nn1184

    • Google Scholar
46. 1. Ramirez S
    2. Liu X
    3. Lin PA
    4. Suh J
    5. Pignatelli M
    6. Redondo RL
    7. Ryan TJ
    8. Tonegawa S

    (2013) Creating a false memory in the hippocampus

    Science 341:387–391.

    https://doi.org/10.1126/science.1239073

    • Google Scholar
47. 1. Rokni U
    2. Steinberg O
    3. Vaadia E
    4. Sompolinsky H

    (2003)

    Cortical representation of bimanual movements

    Journal of Neuroscience 23:11577–11586.

    • Google Scholar
48. 1. Salimpour Y
    2. Shadmehr R

    (2014) Motor costs and the coordination of the two arms

    Journal of Neuroscience 34:1806–1818.

    https://doi.org/10.1523/JNEUROSCI.3095-13.2014

    • Google Scholar
49. 1. Sarwary AM
    2. Selen LP
    3. Medendorp WP

    (2013) Vestibular benefits to task savings in motor adaptation

    Journal of Neurophysiology 110:1269–1277.

    https://doi.org/10.1152/jn.00914.2012

    • Google Scholar
50. 1. Sarwary AM
    2. Stegeman DF
    3. Selen LP
    4. Medendorp WP

    (2015) Generalization and transfer of contextual cues in motor learning

    Journal of Neurophysiology 114:1565–1576.

    https://doi.org/10.1152/jn.00217.2015

    • Google Scholar
51. 1. Schaal S
    2. Sternad D
    3. Osu R
    4. Kawato M

    (2004) Rhythmic arm movement is not discrete

    Nature Neuroscience 7:1136–1143.

    https://doi.org/10.1038/nn1322

    • Google Scholar
52. 1. Scheidt RA
    2. Reinkensmeyer DJ
    3. Conditt MA
    4. Rymer WZ
    5. Mussa-Ivaldi FA

    (2000)

    Persistence of motor adaptation during constrained, multi-joint, arm movements

    Journal of Neurophysiology 84:853–862.

    • Google Scholar
53. 1. Shadmehr R
    2. Mussa-Ivaldi FA

    (1994)

    Adaptive representation of dynamics during learning of a motor task

    Journal of Neuroscience 14:3208–3224.

    • Google Scholar
54. 1. Siebner HR
    2. Lang N
    3. Rizzo V
    4. Nitsche MA
    5. Paulus W
    6. Lemon RN
    7. Rothwell JC

    (2004) Preconditioning of low-frequency repetitive transcranial magnetic stimulation with transcranial direct current stimulation: evidence for homeostatic plasticity in the human motor cortex

    Journal of Neuroscience 24:3379–3385.

    https://doi.org/10.1523/JNEUROSCI.5316-03.2004

    • Google Scholar
55. 1. Tanaka S
    2. Hanakawa T
    3. Honda M
    4. Watanabe K

    (2009) Enhancement of pinch force in the lower leg by anodal transcranial direct current stimulation

    Experimental Brain Research 196:459–465.

    https://doi.org/10.1007/s00221-009-1863-9

    • Google Scholar
56. 1. Yokoi A
    2. Hirashima M
    3. Nozaki D

    (2011) Gain field encoding of the kinematics of both arms in the internal model enables flexible bimanual action

    Journal of Neuroscience 31:17058–17068.

    https://doi.org/10.1523/JNEUROSCI.2982-11.2011

    • Google Scholar
57. 1. Yokoi A
    2. Hirashima M
    3. Nozaki D

    (2014) Lateralized sensitivity of motor memories to the kinematics of the opposite arm reveals functional specialization during bimanual actions

    Journal of Neuroscience 34:9141–9151.

    https://doi.org/10.1523/JNEUROSCI.2694-13.2014

    • Google Scholar

Author details

1. Daichi Nozaki

   Division of Physical and Health Education, Graduate School of Education, The University of Tokyo, Tokyo, Japan

   Contribution

   DN, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   nozaki@p.u-tokyo.ac.jp

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-1338-8337

2. Atsushi Yokoi

   1. The Brain and Mind Institute, University of Western Ontario, London, Canada
   2. Graduate School of Frontier Biosciences, Osaka University, Suita, Japan

   Contribution

   AY, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-7428-3344

3. Takahiro Kimura

   Research Institute, Kochi University of Technology, Kami City, Japan

   Contribution

   TK, Acquisition of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-5673-1553

4. Masaya Hirashima

   Center for Information and Neural Networks, National Institute of Information and Communications Technology, and Osaka University, Suita, Japan

   Contribution

   MH, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0001-8571-8289

5. Jean-Jacques Orban de Xivry

   1. Institute of Information and Communication Technologies, Electronics, and Applied Mathematics, Université catholique de Louvain, Louvain-La-Neuve, Belgium
   2. Institute of Neuroscience, Université catholique de Louvain, Louvain-La-Neuve, Belgium
   3. Department of Kinesiology, Movement Control and Neuroplasticity Research Group, KU Leuven, Leuven, Belgium

   Contribution

   J-JOdX, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-4603-7939

• 4,690

  views

• 801

  downloads

• 22

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.