What type of neurons respond to both self produced movement and an observation of a similar movement by another individual?

J Neurophysiol. 2019 Dec 1; 122(6): 2630–2635.

Abstract

Mirror neurons are thought to represent an individual’s ability to understand the actions of others by discharging as one individual performs or observes another individual performing an action. Studies typically have focused on mirror neuron activity during action observation, examining activity during action execution primarily to validate mirror neuron involvement in the motor act. As a result, little is known about the precise role of mirror neurons during action execution. In this study, during execution of reach-grasp-manipulate movements, we found activity of mirror neurons generally preceded that of non-mirror neurons. Not only did the onset of task-related modulation occur earlier in mirror neurons, but state transitions detected by hidden Markov models also occurred earlier in mirror neuron populations. Our findings suggest that mirror neurons may be at the forefront of action execution.

NEW & NOTEWORTHY Mirror neurons commonly are thought to provide a neural substrate for understanding the actions of others, but mirror neurons also are active during action execution, when additional, non-mirror neurons are active as well. Examining the timing of activity during execution of a naturalistic reach-grasp-manipulate task, we found that mirror neuron activity precedes that of non-mirror neurons at both the unit and the population level. Thus mirror neurons may be at the leading edge of action execution.

Keywords: grasp, hidden Markov models, manipulation, reach

INTRODUCTION

Actions are thought to be elaborated through a neural hierarchy in which a representation of the intention to achieve a particular goal is converted into the necessary patterns of muscle activity and/or movement kinematics (Grafton 2010; Rizzolatti and Luppino 2001). Mirror neurons (MNs) not only participate in this process of action execution but also receive visual information from observation of movements performed by other individuals (Albertini et al. in press; Bruni et al. 2018; di Pellegrino et al. 1992; Dushanova and Donoghue 2010; Gallese et al. 1996; Kraskov et al. 2014; Lanzilotto et al. 2019; Papadourakis and Raos 2019). Their activity during action observation has suggested that MNs contribute to understanding the intentions of others (Kilner and Lemon 2013; Rizzolatti and Fogassi 2014). In contrast, non-mirror neurons (non-MNs) are active only during action execution, presumably representing muscle activity patterns and movement kinematics, but not intentions. If MN activity includes representation of muscle activity and movement kinematics along with representation of one’s own intentions and intermediate goals, then during action execution, MN activity might be hypothesized to modulate before non-MN activity. In this study we explicitly tested the hypothesis that during action execution, the activity of MNs precedes that of non-MNs in both the premotor and motor cortex.

MATERIAL AND METHODS

Animal models.

Two male rhesus macaques (Macaca mulatta; monkeys L and X) were used in the present study. All procedures for the care and use of nonhuman primates followed the Guide for the Care and Use of Laboratory Animals and were approved by the University Committee on Animal Resources at the University of Rochester.

Behavioral task.

As described previously (Mazurek et al. 2018; Mollazadeh et al. 2011), each monkey was trained to sit in a primate chair and perform the reach, grasp, manipulate (RGM) task described below (execution trials, Fig. 1A, left). In addition, each monkey observed an experimenter performing the RGM task (observation trials, Fig. 1A, right). The RGM task apparatus consisted of a “home” object surrounded by four target objects: a perpendicular cylinder, a coaxial cylinder, a button, and a sphere. The target objects were separated by 45° intervals on a 13-cm-radius circle centered on the home object positioned 10 cm from the monkey’s right shoulder.

A: monkeys (execution) or experimenters (observation) performed the reach, grasp, manipulate task. Blue lights instructed the target object. Hand postures in each of the 4 behavioral epochs (labeled at left) have been redrawn from selected video frames. B: neurons were recorded from floating microelectrode arrays implanted in premotor cortex (PM; open squares) and primary motor cortex (M1; closed squares). In monkey L, all PM arrays were situated in the ventral PM, but in monkey X, the more medial PM array may have been in dorsal PM. AS, arcuate sulcus; C, caudal; CS, central sulcus; M, medial.

An execution trial was initiated as the monkey pulled the home object and held it for a variable initial hold period (1.0 to 1.5 s). A ring of blue light-emitting diodes (LEDs) then illuminated around the base of one of the target objects, instructing the monkey to reach to, grasp, and manipulate that object (e.g., pull the perpendicular cylinder, pull the coaxial cylinder, push the button, or turn the sphere). A second ring of green LEDs at the base of each object illuminated when that object was manipulated, as detected by closure of a microswitch. The monkey was required to maintain the manipulated position for a fixed final hold period (0.9 s for monkey L; 1.0 s for monkey X), after which a liquid reward was delivered.

During observation trials, both of the monkey’s arms were constrained in the primate chair as an experimenter standing to the monkey’s right performed the RGM task. The monkey thus had a first person view of the experimenter’s right forearm and hand. The experimenter grasped the home object and, when instructed by the blue LEDs, reached to, grasped, and manipulated the target object, holding the object for the final hold duration, after which the monkey received a liquid reward. To keep the monkey attentive, the experimenter occasionally made errors, after which no reward was delivered. The monkey was allowed to make eye movements ad libitum, and the eye movements were not tracked in any way.

Each monkey performed (execution) and observed the experimenter performing (observation) RGM trials in separate blocks during each recording session. During both execution and observation trials, the target object was selected in a pseudorandom block design. If a trial ended in an error, the subsequent trial involved the same target object until a trial was performed successfully. This prevented the monkey from avoiding trials of particular target objects. Because the monkey therefore could predict the target object in trials following an error, only successful trials not preceded by an error trial were included in subsequent analyses.

The behavioral task was controlled by custom software written in TEMPO (Reflective Computing, Olympia, WA). During each trial, behavioral event marker codes were sent into the recorded data stream to mark the time of specific behavioral events, including instruction onset, movement onset, object manipulation, and final hold period completion. Behavioral epochs in each trial were defined based on these behavioral event markers: initial hold epoch, start of trial to instruction onset; reaction epoch, instruction onset to movement onset; movement epoch, movement onset to object manipulation; and final hold epoch, object manipulation to final hold period completion.

Neural recordings.

Single-unit and multiunit activity were recorded from floating microelectrode arrays (FMAs; MicroProbes for Life Science, Gaithersburg, MD) implanted in primary motor (M1) and premotor (PM) areas as described in detail previously (Mollazadeh et al. 2011). Each FMA had 16 recording electrodes of various lengths (1–9 mm). For the present study, data were analyzed from four FMAs implanted in PM and four FMAs implanted in M1 for monkey L, and from two FMAs implanted in PM and six FMAs implanted in M1 in monkey X (Fig. 1B). In both monkeys, PM electrodes were located in the posterior bank and lip of the arcuate sulcus, with the majority in ventral PM; M1 electrodes were located along the anterior bank and lip of the central sulcus spanning the upper extremity representation.

For each monkey, we recorded spiking neural activity from these FMAs simultaneously in three recording sessions. At the beginning of each session, thresholds were set for spiking activity from each electrode using Sort Client (Plexon, Dallas, TX). Neuron spike times and action potential waveforms were captured using a Multichannel Acquisition Processor (Plexon MAP system) along with the behavioral event markers generated by TEMPO.

Offline, neuron spike times were sorted further using Plexon’s Offline Sorter, followed by a custom sorting algorithm. Each sorted unit was classified as a definite single unit, probable single unit, or multiunit based on the signal-to-noise ratio (SNR) of its waveforms and the fraction of true spikes unlikely to have originated from another neuron or from noise (Hill et al. 2011; Meunier et al. 2003; Rouse and Schieber 2016). Any units with an SNR < 1.5 were discarded. Definite single units, probable single units, and multiunits were all considered “neurons” and included in subsequent analyses.

Each neuron was analyzed for task-related modulation separately during execution and observation trials by using repeated-measures ANOVA (RM-ANOVA) to identify significant temporal variation in firing rates over the course of the trial. For each successful trial of any target object that followed a successful trial, the unit’s spike rate was binned in 100-ms nonoverlapping windows aligned at instruction onset. A repeated-measures model was fit to the data and a two-way ANOVA (time × object) performed. A unit was considered to have modulated significantly during execution or observation if a significant effect of time or of time × object was detected (Bonferroni-corrected for the number of effects, P < 0.05/2 = 0.025). Any unit that modulated significantly for both execution and observation trials was considered an MN; any unit that modulated significantly only for execution trials was considered a non-MN. Note that we used activity not only during the movement epoch of trials, but potentially during the reaction and final hold epochs as well, to classify units as MNs vs. non-MNs (Bonini et al. 2014; Livi et al. 2019).

Modulation onset analyses.

Modulation onset times for each neuron were computed by realigning spike times at movement onset. Perievent histograms were created with 25-ms bins overlapping at 5-ms time steps. Baseline thresholds (mean ± 2 SD) were computed from similarly binned spike times in the same trials but using the neural activity occurring in the 500 ms preceding the instruction onset. Onset times were identified as the time at which activity increased above or decreased below 2 SD from the baseline mean for at least 100 ms averaged across trials involving all four target objects. With the use of this criterion, modulation onsets were not identified for a fraction of neurons (160/530) for which the averaged firing rates did not cross this 2-SD threshold, potentially due to different modulation patterns for each object.

Hidden Markov modeling.

For each of the populations of either MNs or non-MNs recorded in a given session from either PM or M1, hidden Markov models (HMMs) were trained on successful execution trials to detect four hidden states. Use of four hidden states was found previously to provide the most consistent results across trials, sessions, and monkeys. The numbers of neurons in each population from each session have been reported previously (Table 2 in Mazurek et al. 2018). For each trial, a single input observation sequence was created based on the spike times in the simultaneously recorded MN or non-MN population binned in 2-ms nonoverlapping windows. This observation sequence spanned 500 ms before instruction onset to 500 ms after final hold completion. The observation sequence varied in length from trial to trial because the monkeys’ reaction and movement times varied from trial to trial. A unique identifier (ID) ranging from 1 to N, the total number of neurons in the population, was assigned to each bin in the observation sequence. A zero was assigned to bins in which no spike occurred. If a spike from more than one unit in the population occurred in the same bin, then the ID of one of the discharging units selected randomly was assigned to that bin. Such “collisions” occurred in a median of only 3.2% of bins across all models trained. Sacrificing this small fraction of information about individual neuron spikes permitted the observation sequence describing the entire population’s activity in any given trial to consist of a one-dimensional vector of neuron IDs and zeroes. These observation sequences were used to estimate the HMM parameters: a transition matrix (the probabilities of a state transition occurring between any pair of the 4 states) and an emission matrix (the probabilities that a unit will fire in each of the states). HMM parameters were estimated recursively using the Baum–Welch algorithm (Baum et al. 1970) until the log-likelihood converged to less than a tolerance factor (<10−6) or for a maximum of 500 iterations. Because the Baum–Welch algorithm guaranteed approaching a local maximum but not necessarily a global maximum, each model was trained from 10 different initial conditions to improve the chances of finding the global maximum. The model resulting from these 10 initial conditions with the greatest log-likelihood was selected for subsequent analysis. Note that the training of HMMs incorporated no information about behavioral event times.

HMM state analyses.

After HMMs had been trained on execution trials separately for each of the four objects, we identified sequences of states for each trial (Fig. 2A). A state was considered active if the state probability exceeded an arbitrarily selected threshold of 0.6, and the time at which the state first exceeded a probability of 0.6 was considered its rise time. In previous work, we analyzed the consistency of hidden Markov states in the present data set (Figs. 4–6 of (Mazurek et al. 2018). We found that in most trials, HMMs detected a consistent ordinal sequence of four hidden states corresponding approximately to the four behavioral epochs present in every trial. We refer to each hidden state according to the behavioral epoch during which that state first became active, on average, across trials. For each of four neural population types (PM MNs, PM non-MNs, M1 MNs, and M1 non-MNs), we identified trials in which the sequence included all four hidden states in order: initial, reaction, movement, and final (numbers in Table 1). Outlier trials in which any of the three rise-time latencies for either population was more than 3 SD from its mean were excluded from further analysis (Table 1). For the present pairwise comparisons, we selected trials in which all four hidden states occurred in order in each of two of the population types (Table 1). In each of these trials, for each population type, we measured the rise-time latency of the reaction, movement, and final states relative to the behavioral event markers that indicated the beginning of the corresponding behavioral epochs in that trial: the reaction state relative to instruction onset, the movement state relative to movement onset, and the final state relative to the start of the final hold.

A: hidden state sequences detected by hidden Markov models in a single trial. Colored traces representing the probability of 4 hidden states [initial (blue), reaction (red), movement (green), and final (purple)] overlie the rastered activity of simultaneously recorded populations of premotor cortex (PM) mirror neurons (MNs; top) and PM non-MNs (bottom). Black markers below raster plots indicate the time of instruction onset (square), movement onset (circle), and object manipulation (triangle). Colored dashed and dotted vertical lines indicate the times at which the reaction, movement, and final states exceeded the 0.6 probability threshold (horizontal gray lines) used to define rise times for each population, with corresponding colored triangles indicating the duration between state transitions in the MN vs. non-MN population from PM. B: similar to A for primary motor cortex (M1) MNs (top) and M1 non-MNs (bottom). C: perievent histograms and multiple trial spike rasters for a single PM MN (left) and a single PM non-MN (right). Top frames show firing rates in 25-ms overlapping bins stepped at 5-ms intervals averaged across all trials involving each object separately (colored traces) and all trials together (solid black line); total number of trials (nTrials) is indicated above frames. Modulation onset time, indicated by downward arrows, was defined when the firing rate averaged across all trials (black trace) crossed ±2 SD (dotted horizontal lines) from the mean baseline activity (horizontal gray lines) measured 500 ms preceding instruction onset). Bottom frames show spike rasters for 5 trials involving each object. Traces involving each object are colored differently [sphere (dark blue), button (orange), coaxial cylinder (teal), and perpendicular cylinder (pink)]. Black markers indicate the same behavioral events as in A and B, with inverted triangles indicating the end of the final hold period. D: similar to C for a single M1 MN (left) and a single M1 non-MN (right).

Table 1.

Summary of number of trials collected and analyzed

Statistical methods.

Because the distributions of rise-time latencies and modulation onset times were not necessarily normally distributed, measures for the different neural populations were compared nonparametrically using Wilcoxon signed-rank tests. Uncorrected P values are reported throughout.

RESULTS

Hidden state transitions in MN populations precede those in non-MN populations.

We described previously that HMMs trained on the activity of MN or non-MN populations in PM or M1 during execution trials detect a sequence of four hidden states corresponding approximately to the four behavioral epochs of the RGM task (Mazurek et al. 2018). Figure 2A shows rastered spike times from an example trial in which 30 MNs and 22 non-MNs were recorded simultaneously from PM, and Fig. 2B depicts another example trial in which 28 MNs and 37 non-MNs were recorded simultaneously from M1. For each neural population, an initial state transitioned to a reaction state during the reaction time, followed by a transition to a movement state during the movement time and then a final state during the final hold period. The sequence of four hidden states through which these neural populations progressed was consistent across multiple trials both for MN and for non-MN populations both in PM and in M1.

Previously, we had found that, averaged across all trials, these hidden state transitions occurred earlier in MNs than in non-MNs, albeit with considerable overlap. In the present study, we further tested the hypothesis that MN populations precede non-MN populations during action execution, now using paired comparisons in single trials from the same data sets reported previously (Mazurek et al. 2018). We first identified individual RGM trials in which all four hidden states were detected in both a MN population and in the simultaneously recorded non-MN population in either PM or M1. Table 1 gives the numbers of trials accepted and the number rejected as outliers (either for showing fewer than 4 states or for having the 4 states in an atypical order). We defined the time of each state transition when the probability of the preceding state had fallen below and the current state then rose above an arbitrarily chosen probability of 0.6.

In the example trials of Fig. 2, A and B, vertical lines mark these rise times of the reaction, movement, and final states in the MN population (dashed) and in the simultaneously recorded non-MN population (dotted). Both sets of vertical lines have been projected down, and triangles between them indicate the duration from the rise time of a given state in the MN population to the corresponding rise time in the non-MN population. For all the transitions in these two example trials, the MN population led the simultaneously recorded non-MN population, albeit by varying time differences.

To compare the timing of these state transitions across trials, for each trial in which all four hidden states were detected by a given population we calculated 1) the rise-time latency of the reaction state following instruction onset, 2) the rise-time latency of the movement state following movement onset, and 3) the rise-time latency of the final state following completion of the manipulation starting the final hold. We then identified trials in which these latencies had been measured for two different types of simultaneously recorded populations and used those trials to compare rise-time latencies between the populations: PM MNs vs. PM non-MNs and M1 MNs vs. M1 non-MNs. For each comparison, we formed a scatterplot in which each point represented the rise-time latency for one population plotted against the corresponding rise-time latency for the other population in a given trial, pooling data across transitions to the reaction, movement, and final states; across trials involving all four objects; and across all three sessions from both monkeys. Such plots for MN vs. non-MN populations are shown for PM populations in Fig. 3A and for M1 populations in Fig. 3B. The majority of points in each scatterplot fell above the dotted line of unity slope, indicating the rise-time latency of the MN population most often was shorter than that of the non-MN population both in PM (P = 7.1e-12, Wilcoxon signed-rank test) and in M1 (P = 1.3e-9).

Comparisons of hidden state rise-time latencies and modulation onset times between mirror neurons (MNs) and non-MNs in premotor (PM) and primary motor cortex (M1). Scatterplots show hidden state rise-time latencies in PM MN vs. PM non-MN populations (A) and in M1 MN vs. M1 non-MN populations (B). A diagonal dotted line of unity slope in each plot divides transitions that occurred earlier in the former population (points above and left) from transitions that occurred earlier in the latter population (points below and right). Colored traces represent reaction (red), movement (green), and final (blue) state transitions. Modulation onset time histograms compare PM MN (blue) vs. non-MN (red) populations (C) and M1 MN (orange) vs. M1 non-MN (green) populations (D), demonstrating that individual MN onset times preceded those of non-MNs. Only trials in which the hidden Markov models detected 4 hidden states were used for computing these onset time distributions. Numbers of neurons in which onset times were detected are displayed in plot insets. Distributions were compared using Wilcoxon rank sum test. All P values were significant after Bonferroni correction for 2 comparisons.

Modulation began earlier in MNs than in non-MNs.

The initial transitions to the reaction state were detected in simultaneously recorded populations of neurons. Did the timing of modulation onset in individual neurons differ as well? To address this question, we compared the modulation onset times of MNs vs. non-MNs in PM and in M1. For each MN or non-MN, we created a perievent time histogram aligned at movement onset, using all trials for which the neural population progressed through the consistent sequence of four states detected by the HMM. We then defined the onset time of each neuron’s modulation as having occurred when the spike count in 25-ms overlapping bins deviated more than ±2 SD from the baseline period (500 ms preceding instruction onset) for at least 100 ms. We assigned negative values if modulation onset occurred before movement onset and positive values if modulation onset followed movement onset. Figure 2C depicts spike rasters and firing rates from an MN and from a non-MN both recorded in PM, each of which modulated significantly during execution of RGM movements. Onset times, indicated by downward arrows, were −145 ms (before) movement onset for the MN and +25 ms (after) movement onset for the non-MN. Figure 2D depicts similar examples from M1 in which onset times were −180 ms (before) movement onset for the MN and −115 ms (before) movement onset for the non-MN.

As illustrated in Fig. 3C, modulation onset times for PM MNs (median, −15 ms) were significantly earlier on average than those of non-MNs (median, +72.5 ms; P = 0.023, Wilcoxon rank sum test). In M1 as well (Fig. 3D), MNs had modulation onset times (median, −170 ms) that were on average earlier than those of non-MNs (median, −105 ms; P = 1.2e-6). Not only did hidden state transitions occur earlier on average in MN populations than in non-MN populations, but modulation onset was earlier in individual MNs than in non-MNs.

DISCUSSION

Perhaps the most prominent feature of MNs is their activity during action observation, when many other neurons in the premotor and primary motor cortex are inactive. During action observation, MNs are thought to monitor the movements of other individuals to understand the intention behind the action being performed (di Pellegrino et al. 1992; Gallese et al. 1996; Rizzolatti and Craighero 2004; Rizzolatti et al. 1996, 2001; Umiltà et al. 2001). Relatively few studies have investigated the role of MNs during action execution, when the activity of MNs is generally considered to be similar to that of non-MNs.

Although MNs are generally thought to have similar activity during action execution and action observation, the majority of MNs actually show incongruent patterns of modulation (Breveglieri et al. 2019; Ferrari et al. 2003; Mazurek et al. 2018; Papadourakis and Raos 2019), with many MNs showing a less dramatic increase in firing rate during observation than during execution, or even a suppression of firing rate during observation (Kraskov et al. 2009; Vigneswaran et al. 2013). MNs constitute a substantial fraction of task-related neurons in both PM and M1, and a considerable number of MNs in each region have corticospinal axons. Less corticospinal drive during action observation then may serve to prevent self-movement (Kraskov et al. 2014). However, participation of MNs in driving limb movement during action execution along with non-MNs still does not account for the average temporal lead of MNs.

Other studies have suggested that whereas non-MNs drive movement execution, MNs monitor the self-movement, thereby confirming the appropriateness of one’s own actions (Maranesi et al. 2015). If MNs received only reafferent feedback from the moving limb (proprioceptive or visual) and/or efference copy from non-MNs, we would expect the activity of non-MNs to be underway before that of MNs. Thus our findings suggest that monitoring self-movement may not be the only role of MNs during action execution, and observational visual information not their only sensory input.

Our findings suggest instead that MNs might form the leading edge of the overall population of neurons involved in executing movements. In preceding non-MNs, MNs may provide an overall prediction about the goal of the action being performed (Gallese et al. 1996) and/or a stepwise prediction about the sequence of events needed to achieve the goal (e.g., once instruction has been received, initiate movement; once movement has been initiated, prepare to end movement; once movement has ended, continue to hold). MNs may be able to play a predictive role during action execution if they receive a larger set of sensory and contextual information [e.g., the nature of the target object and how it is to be used (Bonini et al. 2010; Livi et al. 2019)] than non-MNs. Further studies will be needed to determine whether MNs use such information predictively to lead non-MNs in the process of action execution.

GRANTS

This work was supported by National Institute of Neurological Disorders and Stroke Grants F32NS093709 (to K. A. Mazurek) and R01NS079664 and R01NS102343 (to M. H. Schieber).

DISCLAIMERS

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

K.A.M. and M.H.S. conceived and designed research; M.H.S. performed experiments; K.A.M. analyzed data; K.A.M. and M.H.S. interpreted results of experiments; K.A.M. prepared figures; K.A.M. drafted manuscript; K.A.M. and M.H.S. edited and revised manuscript; K.A.M. and M.H.S. approved final version of manuscript.

ACKNOWLEDGMENTS

We thank Marsha Hayles for editorial comments.

REFERENCES

• Albertini D, Gerbella M, Lanzilotto M, Livi A, Maranesi M, Ferroni CG, Bonini L. Connectional gradients underlie functional transitions in monkey pre-supplementary motor area. Prog Neurobiol. In press. doi: 10.1016/j.pneurobio.2019.101699. [PubMed] [CrossRef] [Google Scholar]
• Baum LE, Petrie T, Soules G, Weiss N. A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains. Ann Math Stat 41: 164–171, 1970. doi: 10.1214/aoms/1177697196. [CrossRef] [Google Scholar]
• Bonini L, Maranesi M, Livi A, Fogassi L, Rizzolatti G. Ventral premotor neurons encoding representations of action during self and others’ inaction. Curr Biol 24: 1611–1614, 2014. doi: 10.1016/j.cub.2014.05.047. [PubMed] [CrossRef] [Google Scholar]
• Bonini L, Rozzi S, Serventi FU, Simone L, Ferrari PF, Fogassi L. Ventral premotor and inferior parietal cortices make distinct contribution to action organization and intention understanding. Cereb Cortex 20: 1372–1385, 2010. doi: 10.1093/cercor/bhp200. [PubMed] [CrossRef] [Google Scholar]
• Breveglieri R, Vaccari FE, Bosco A, Gamberini M, Fattori P, Galletti C. Neurons modulated by action execution and observation in the macaque medial parietal cortex. Curr Biol 29: 1218–1225.e3, 2019. doi: 10.1016/j.cub.2019.02.027. [PubMed] [CrossRef] [Google Scholar]
• Bruni S, Gerbella M, Bonini L, Borra E, Coudé G, Ferrari PF, Fogassi L, Maranesi M, Rodà F, Simone L, Serventi FU, Rozzi S. Cortical and subcortical connections of parietal and premotor nodes of the monkey hand mirror neuron network. Brain Struct Funct 223: 1713–1729, 2018. doi: 10.1007/s00429-017-1582-0. [PubMed] [CrossRef] [Google Scholar]
• di Pellegrino G, Fadiga L, Fogassi L, Gallese V, Rizzolatti G. Understanding motor events: a neurophysiological study. Exp Brain Res 91: 176–180, 1992. doi: 10.1007/BF00230027. [PubMed] [CrossRef] [Google Scholar]
• Dushanova J, Donoghue J. Neurons in primary motor cortex engaged during action observation. Eur J Neurosci 31: 386–398, 2010. doi: 10.1111/j.1460-9568.2009.07067.x. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Ferrari PF, Gallese V, Rizzolatti G, Fogassi L. Mirror neurons responding to the observation of ingestive and communicative mouth actions in the monkey ventral premotor cortex. Eur J Neurosci 17: 1703–1714, 2003. doi: 10.1046/j.1460-9568.2003.02601.x. [PubMed] [CrossRef] [Google Scholar]
• Gallese V, Fadiga L, Fogassi L, Rizzolatti G. Action recognition in the premotor cortex. Brain 119: 593–609, 1996. doi: 10.1093/brain/119.2.593. [PubMed] [CrossRef] [Google Scholar]
• Grafton ST. The cognitive neuroscience of prehension: recent developments. Exp Brain Res 204: 475–491, 2010. doi: 10.1007/s00221-010-2315-2. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Hill DN, Mehta SB, Kleinfeld D. Quality metrics to accompany spike sorting of extracellular signals. J Neurosci 31: 8699–8705, 2011. doi: 10.1523/JNEUROSCI.0971-11.2011. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Kilner JM, Lemon RN. What we know currently about mirror neurons. Curr Biol 23: R1057–R1062, 2013. doi: 10.1016/j.cub.2013.10.051. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Kraskov A, Dancause N, Quallo MM, Shepherd S, Lemon RN. Corticospinal neurons in macaque ventral premotor cortex with mirror properties: a potential mechanism for action suppression? Neuron 64: 922–930, 2009. doi: 10.1016/j.neuron.2009.12.010. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Kraskov A, Philipp R, Waldert S, Vigneswaran G, Quallo MM, Lemon RN. Corticospinal mirror neurons. Philos Trans R Soc Lond B Biol Sci 369: 20130174, 2014. doi: 10.1098/rstb.2013.0174. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Lanzilotto M, Ferroni CG, Livi A, Gerbella M, Maranesi M, Borra E, Passarelli L, Gamberini M, Fogassi L, Bonini L, Orban GA. Anterior intraparietal area: a hub in the observed manipulative action network. Cereb Cortex 29: 1816–1833, 2019. doi: 10.1093/cercor/bhz011. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Livi A, Lanzilotto M, Maranesi M, Fogassi L, Rizzolatti G, Bonini L. Agent-based representations of objects and actions in the monkey pre-supplementary motor area. Proc Natl Acad Sci USA 116: 2691–2700, 2019. doi: 10.1073/pnas.1810890116. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Maranesi M, Livi A, Bonini L. Processing of own hand visual feedback during object grasping in ventral premotor mirror neurons. J Neurosci 35: 11824–11829, 2015. doi: 10.1523/JNEUROSCI.0301-15.2015. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Mazurek KA, Rouse AG, Schieber MH. Mirror neuron populations represent sequences of behavioral epochs during both execution and observation. J Neurosci 38: 4441–4455, 2018. doi: 10.1523/JNEUROSCI.3481-17.2018. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Meunier N, Marion-Poll F, Lansky P, Rospars JP. Estimation of the individual firing frequencies of two neurons recorded with a single electrode. Chem Senses 28: 671–679, 2003. doi: 10.1093/chemse/bjg059. [PubMed] [CrossRef] [Google Scholar]
• Mollazadeh M, Aggarwal V, Davidson AG, Law AJ, Thakor NV, Schieber MH. Spatiotemporal variation of multiple neurophysiological signals in the primary motor cortex during dexterous reach-to-grasp movements. J Neurosci 31: 15531–15543, 2011. doi: 10.1523/JNEUROSCI.2999-11.2011. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Papadourakis V, Raos V. Neurons in the macaque dorsal premotor cortex respond to execution and observation of actions. Cereb Cortex 29: 4223–4237, 2019. doi: 10.1093/cercor/bhy304. [PubMed] [CrossRef] [Google Scholar]
• Rizzolatti G, Craighero L. The mirror-neuron system. Annu Rev Neurosci 27: 169–192, 2004. doi: 10.1146/annurev.neuro.27.070203.144230. [PubMed] [CrossRef] [Google Scholar]
• Rizzolatti G, Fadiga L, Gallese V, Fogassi L. Premotor cortex and the recognition of motor actions. Brain Res Cogn Brain Res 3: 131–141, 1996. doi: 10.1016/0926-6410(95)00038-0. [PubMed] [CrossRef] [Google Scholar]
• Rizzolatti G, Fogassi L. The mirror mechanism: recent findings and perspectives. Philos Trans R Soc Lond B Biol Sci 369: 20130420, 2014. doi: 10.1098/rstb.2013.0420. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Rizzolatti G, Fogassi L, Gallese V. Neurophysiological mechanisms underlying the understanding and imitation of action. Nat Rev Neurosci 2: 661–670, 2001. doi: 10.1038/35090060. [PubMed] [CrossRef] [Google Scholar]
• Rizzolatti G, Luppino G. The cortical motor system. Neuron 31: 889–901, 2001. doi: 10.1016/S0896-6273(01)00423-8. [PubMed] [CrossRef] [Google Scholar]
• Rouse AG, Schieber MH. Spatiotemporal distribution of location and object effects in primary motor cortex neurons during reach-to-grasp. J Neurosci 36: 10640–10653, 2016. doi: 10.1523/JNEUROSCI.1716-16.2016. [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Umiltà MA, Kohler E, Gallese V, Fogassi L, Fadiga L, Keysers C, Rizzolatti G. I know what you are doing. a neurophysiological study. Neuron 31: 155–165, 2001. doi: 10.1016/S0896-6273(01)00337-3. [PubMed] [CrossRef] [Google Scholar]
• Vigneswaran G, Philipp R, Lemon RN, Kraskov A. M1 corticospinal mirror neurons and their role in movement suppression during action observation. Curr Biol 23: 236–243, 2013. doi: 10.1016/j.cub.2012.12.006. [PMC free article] [PubMed] [CrossRef] [Google Scholar]

Articles from Journal of Neurophysiology are provided here courtesy of American Physiological Society