Motor (M1) and dorsal premotor (PMd) cortex in monkeys are widely used as models for studying cortical control of movement. Both cortical areas contain complete motor representations of the forelimb (e.g., Boudrias et al., 2010; Gould et al., 1986; Murphy et al., 1978; Raos et al., 2003). Somatotopy of the forelimb representations varies across studies, but there is consensus that muscles and cortical columns do not have a one-to-one mapping. Instead, an arm or a hand muscle receives corticospinal inputs from a population of cortical columns (Andersen et al., 1975; Rathelot and Strick, 2006). Similarly, a cortical column can influence activity in several muscles (Donoghue et al., 1992; Fetz and Cheney, 1980; Lemon et al., 1987). The convergence and divergence of corticospinal projections mean that the same motor output is present in many locations within a forelimb representation. For example, intracortical microstimulation (ICMS) evokes shoulder flexion from many M1 sites where the affiliate corticospinal projections target the same group of arm muscles (Park et al., 2001). But how can a seemingly simple motor map control a rich repertoire of arm and hand actions?

To address this question, we must first understand the spatial relationship between neural activity that supports arm and hand actions (i.e., function), and the forelimb motor maps (i.e., structure). We consider two different perspectives. The first perspective is that neural codes for arm and hand actions are not spatially organized within the forelimb motor maps. Instead, neural activity affiliated with a particular function (e.g., reach) is present throughout the arm zones, or even throughout entire forelimb representations. This scheme has support in electrophysiological studies that examined relationships between neural coding, recording site location, and forelimb behavior (Rouse and Schieber, 2016; Saleh et al., 2012; Vargas-Irwin et al., 2010). Nevertheless, most electrophysiological investigations pool results across recording sites, offering only limited insight into the spatial organization of functions. An alternative perspective is that neural codes for arm and hand actions are more spatially organized within the forelimb motor maps. Here, neural activity affiliated with a particular action (e.g., reach) is concentrated in subzones within the forelimb representations. This spatial organization of function is consistent with maps obtained using long train ICMS (500 ms), which approximates the duration of natural actions (Graziano et al., 2002). The same motor mapping parameters revealed functional subzones in both frontal and parietal cortical areas (Cooke and Graziano, 2004; Gharbawie et al., 2011; Graziano et al., 2002; Stepniewska et al., 2005).

Adjudicating between the two perspectives can be facilitated with neuroimaging because it provides uninterrupted spatial sampling and retains the spatial dimension of the recorded activity. The necessary implementation here is to image entire forelimb representations during arm and hand actions then quantify the spatial organization of movement-related cortical activity. Intrinsic signal optical imaging (ISOI), two-photon imaging, and functional magnetic resonance imaging (fMRI) have been successfully used in measuring cortical activity related to arm and hand actions in monkeys (Ebina et al., 2018; Friedman et al., 2020a; Kondo et al., 2018; Nelissen and Vanduffel, 2011). The mesoscopic field-of-view (FOV) in ISOI is particularly well suited for the size of the forelimb representations in M1 and PMd. Moreover, ISOI affords high spatial resolution and contrast without extrinsic agents (e.g., GCaMP, dyes, or MION). These reasons motivated us to use ISOI to measure M1 and PMd activity in head-fixed macaques engaged in an instructed reach-to-grasp task (Figure 1A–D). In the same FOV, we used ICMS for high-density mapping of the forelimb representations and surrounding territories. We then quantified the spatial overlap between cortical activity determined from ISOI and the forelimb motor map. We reasoned that if the neural code for reaching and grasping is not spatially organized within the forelimb representation, then ISOI should report diffuse task-related activity that overlaps most of the M1 and PMd forelimb representations (Figure 1B, top). Alternatively, if the neural code is more spatially organized, then ISOI should report focal task-related activity in subzones of the M1 and PMd forelimb representations (Figure 1B, bottom).

Figure 1

Download asset Open asset

Two monkeys performed an instructed forelimb task that consisted of four conditions: (1) reach-to-grasp with precision grip, (2) reach-to-grasp with power grip, (3) reach-only, and (4) withhold (Figure 1D). Both monkeys completed 49 successful trials/condition/session (median; interquartile range [IQR] = 37–51 trials [monkey G], IQR = 45–51 trials [monkey S]). We used ISOI to measure cortical activity during task performance (Figure 1). The spatial extent of activity in M1 and PMd cortex was quantified in relation to motor maps derived with ICMS from the same recording chambers. Measurements of joint angles and electromyography (EMG) activity during the task provided context for condition differences in cortical activity.

Consistent somatotopy in M1 and PMd

To find the forelimb representations in M1 and PMd, we used ICMS to map frontal cortex from central sulcus to arcuate sulcus (Figure 2A, E). The general organization of the motor map was consistent between monkeys (Figure 2B–D, F–H). Most ICMS sites evoked forelimb movements (Figure 2B, F). We marked the rostral border of M1 such that (1) it was 3–5 mm from the central sulcus, and (2) separated high threshold sites (>30 μA) from low threshold sites (Figure 2C, G). To simplify the motor map, site classifications were consolidated into broader categories (e.g., elbow and shoulder became arm). The simplified maps (Figure 2D, H) showed that the forelimb representations were flanked medially by trunk zones and laterally by upper trunk and face zones. Within the M1 forelimb representation, the main hand zone (i.e., digits and wrist) was surrounded by an arm zone, or an arm and trunk zone. This nested organization is consistent with previous maps from macaque monkeys (Murphy et al., 1978; Sessle and Wiesendanger, 1982) including maps obtained with stimulus triggered averaging of EMG activity (Park et al., 2001). The organization of the forelimb representations was less clear in PMd and PMv, which could have been related to higher thresholds, less extensive mapping, as well as more overlap between arm and hand zones than in M1 (Boudrias et al., 2010).

Figure 2

Download asset Open asset

Movement-related activity in M1 and PMd

Neural activity drives a hemodynamic response that is detectable as reflectance change in ISOI. Under red illumination (630 nm wavelength), which was used here, negative reflectance (i.e., pixel darkening) is a lagging indicator of increased neural activity (Figure 1C). Thus, in the movement conditions, we assumed that pixels would darken in locations where neural activity increased for movement execution, movement planning, or both. We refer to a spatial cluster of darkened pixels as an active ‘patch’, which is conceptually similar to an active ‘domain’, referenced in other studies (e.g., Bonhoeffer and Grinvald, 1991; Lu and Roe, 2007). A domain, however, is typically identified from the differential cortical response to orthogonal conditions (e.g., orientation domains in V1). If task-related neural activity increases throughout the forelimb representations, then we would expect a large patch to fill most of the FOV (Figure 1B, top). If neural activity is more spatially confined, however, then we would expect multiple, smaller, patches (Figure 1B, bottom).

First, we examined reflectance change in M1 and PMd in an average time series from a representative session (36 trials/condition). In the power grip condition (Figure 3A), there was no reflectance change from baseline to movement onset (−1.0 to +0.5 s from Cue). During reach, grasp, and hold (+0.5 to +2.0 s from Cue), pixels in the center of the FOV brightened (cool colors). In the same period, pixel darkening mostly overlapped the central sulcus and superior arcuate sulcus (landmarks on first panel). At first glance, these observations were counterintuitive as we expected pixel darkening (warm colors) in the center of FOV where neural activity presumably increased in support of arm and hand actions. We will explore the time course of reflectance change in subsequent panels and figures. For now, however, we interpret pixel darkening in this period as indication of activity in major vessels. Once the hand returned to the start position after movement completion (i.e., hold start), we observed the expected reflectance change: pixels in the center of the FOV started to darken and form patches in M1 and PMd (+3.5 s from Cue). Patches continued to darken and expand in the remaining frames. Nevertheless, even at peak intensity and size (+5.0 to +6.0 s from Cue), patches remained spatially separable within M1 and between M1 and PMd. For contrast, Figure 3B shows an average time series from the withhold condition where the Cue instructed the animal to hold the start position for the duration of the trial. Here, pixel darkening was limited to major blood vessels as most of the FOV was unchanged or brightened (Figure 3B, cyan pixels). Thus, the patches in Figure 3A were likely lagging indicators of movement-related neural activity.

Figure 3 with 1 supplement see all

Download asset Open asset

To summarize the activity patterns in the time series, we averaged frames from two phases. For the baseline phase, we averaged 10 frames acquired −1.0 to 0 s from Cue. For the movement phase, we averaged 33 frames captured post-movement (+2.8 to +6.0 s from Cue). Average baseline frames showed no reflectance change and were indistinguishable between the movement and withhold conditions (Figure 3C, D). In contrast, reflectance change was apparent in the movement frame (Figure 3E), but not in the matching period from the withhold condition (Figure 3F). To threshold Figure 3E, we flagged pixels that darkened significantly in the post-movement frame as compared to the baseline frame (Figure 3G, t-test, left tail, p < 0.0001). Thus, red pixels in Figure 3G show locations of movement-related neural activity from a single session. We thresholded Figure 3F for contrast, but to capture the predominant reflectance change here, we focused on pixels that brightened in the temporally matched frame as compared to the baseline frame (Figure 3H, t-test, right tail, p < 0.0001). The small number of zones and their small size indicates that reflectance change was limited in the absence of movement.

Next, we examined the time course of reflectance change in both conditions (Figure 3I, J). We placed regions-of-interest (ROIs) with two guiding objectives (Figure 3G, colored circles). First, to target M1 hand, M1 arm, and PMd arm, which we achieved by consulting the motor maps and blood vessel patterns. Second, to target locations that showed activity in time series averaged across trials from all imaging sessions (e.g., Figure 5A). Time courses from two of the ROIs showed an abrupt, yet small, reflectance change with movement onset (+0.5 s from Cue). This brief change was likely due to motion artifact and will be more apparent in the next figures. After the artifact, the arm ROIs and the hand ROI had distinct time course profiles (Figure 3I). For the arm ROIs, positive reflectance started to increase ~0.3 s after movement onset (+0.8 s from Cue) and peaked during the hold phase (+1.7 s from Cue) when the object was grasped and maintained in the lifted position. Then, negative reflectance increased and peaked 2–3 s after movement completion (+5.0 to +6.0 s from Cue). The positive–negative sequence is discussed later in relation to triphasic time courses (negative–positive–negative) established for sensory cortex (Chen-Bee et al., 2007; Sirotin and Das, 2009). In contrast, the time course of the hand ROI did not show the positive reflectance that lasted for several seconds in the arm ROIs. Instead, the time course was flat until negative reflectance began to increase +2.5 s from Cue. The timing of the negative peak was consistent, however, across the hand and arm ROIs. The distinctiveness of the time courses from the arm and hand may reflect functional differences between those zones. In contrast to the profiles in Figure 3I, the same ROIs showed no reflectance change in the withhold condition (Figure 3J).

Pixel darkening in M1 and PMd is locked to movement

Our next objective was to determine whether movement execution was necessary for the pixel darkening in the movement condition. We addressed this point in three separate experiments. In Experiment 1, one monkey performed reach-to-grasp trials interleaved with no-go trials. In a no-go trial, the Go Cue and preceding steps were no different from a trial in a movement condition. In a no-go trial, however, 250 ms after the Go Cue (i.e., during reaction time), another Cue (LED plus tone) instructed the monkey to hold the start position instead of move. Correct trials were typically achieved with the monkey releasing the start position to reach and then immediately returning its hand to the start position. Movements in no-go trials were therefore truncated but not inhibited. If the monkey reached past the midpoint between the start position and the target, then the trial was considered incorrect. We reasoned that no-go trials would minimize movement-related cortical activity without interfering with internal processes that precede movement (e.g., movement preparation). Results were obtained from two imaging sessions. In the movement condition, the thresholded map showed patches of activity in post-movement frames as compared to baseline frames (Figure 3K, t-test, p < 0.0001). In contrast, the thresholded map from the no-go condition had fewer and smaller patches (Figure 3L). Experiment 1 therefore shows that significant pixel darkening was contingent on movement execution.

In Experiment 2, we added conditions in which the monkey did not move but observed an experimenter perform the task instead. First, the monkey completed blocks of trials with regular movement conditions (performed trials). Next, the monkey observed the experimenter perform blocks of trials of the same movement conditions (observed trials). The primate chair was closed off in the observed trials to prevent the monkey from reaching. Cues, timing, and reward schedule were consistent between performed and observed trials. Time courses were measured from ROIs in M1 and PMd (Figure 3—figure supplement 1A). In M1, time courses had a clear negative peak in the performed trials (Figure 3—figure supplement 1B) but remained near baseline in the observed trials. In PMd however, the performed and observed trials had overlapping time courses. All negative peaks in PMd were smaller than the M1 peaks in the performed trials. Results from Experiment 2 suggest that M1 activity was driven by movement execution, whereas PMd activity was likely related to movement execution, movement preparation, and motor cognition.

In Experiment 3, we systematically varied the timing of the Go Cue so that it was 1, 2, or 3 s from trial initiation. We reasoned that temporal shifts in movement onset would lead to predictable shifts in pixel darkening. Only one movement condition was tested in this experiment. We placed ROIs in the M1 hand zone (insets in Figure 4A, B). Time courses from both monkeys confirmed that the negative peaks were locked to Cue onset, and by extension movement onset. This temporal relationship was also evident in the spatial development of cortical activity (Figure 4C). For each Cue condition, we generated thresholded maps from three time windows near the negative peaks. Each thresholded map therefore reported on pixels that darkened in a specific time window as compared to the baseline frame (Figure 4, t-test, p < 0.0001, n = 44 trials). Across Cue conditions, the largest maps were in the time window +5.8 to +6.0 s from Cue (Figure 4C, outlined panels). Thus, Experiment 3 shows that peak map size, and the timing of the negative peak, were both locked to movement onset. Collectively, Experiments 1–3 confirm that the M1 activity reported here was linked to movement execution, whereas PMd activity was likely more complex.

Figure 4

Download asset Open asset

Average time series reveal consistent features of M1 and PMd activity

To identify the most consistent activity patterns, we generated an average time series for each condition. Figure 5A summarizes the average time series for the precision grip condition (monkey G, 8 sessions, 404 trials). Animating the time series side-by-side with a representative behavioral trial (Video 1) qualitatively revealed spatiotemporal features of the cortical activity. (1) During movement, pixel darkening was limited to major vessels; the rest of the FOV brightened or was unchanged. (2) Pixel darkening in cortex (i.e., beyond major vessels) occurred after movement completion. (3) The intensity of pixel darkening and the number of pixels that darkened progressed gradually from movement completion until the end of image acquisition. The same features were present in the other movement conditions (not shown).

Figure 5 with 2 supplements see all

Download asset Open asset

Video 1

Download asset

From the average time series, we condensed the spatiotemporal organization of cortical activity into a single frame. To that end, we used the time course of every pixel to visualize the time of the negative peak (i.e., maximal darkening). Figure 5B color codes peak times from the time series in Figure 5A but omits information related to magnitude of reflectance. Peak times fit into three broad categories (Figure 5B). (1) Early peaks. Green pixels peaked ~2 s from Cue when the monkey was still engaged in task-related movements. Green pixels generally corresponded with the hot color pixels in Figure 5A panels 1.6–2.5, and therefore may have been affiliated with the early response in major vessels. (2) Intermediate peaks. Orange/yellow pixels peaked ~3.5 s from Cue, which was within 1 s of movement completion. Orange/yellow pixels were in the approximate location of yellow pixels in Figure 5A last panel. (3) Late peaks. Magenta/red pixels peaked ~5.5 s from Cue, which was ~3.0 s from movement completion. These pixels corresponded with the hot color pixels in Figure 5A last three panels. Late peak pixels were in clusters surrounded by intermediate peak pixels, but there was no regional divide between the two.

For another perspective on the spatiotemporal patterns of cortical activity, we examined time courses from ROIs that we placed in M1 and PMd (Figure 5B). The three circular ROIs matched those in Figure 3G and returned time courses consistent with Figure 3I, J. In the movement conditions, time courses from the three ROIs had motion artifact that started with reach onset (Figure 5C–E; ~0.5 s from Cue) and lasted for ~0.2 s. For the M1 hand ROI (Figure 5C), after the artifact, reflectance change was flat for ~1.5 s. Negative reflectance started to increase ~2.5 s from Cue, which coincided with the end of movement in the precision and power conditions. In those conditions, the negative peak was larger, and occurred later, than in the reach-only condition. This temporal difference could have been due to the larger size of the negative peaks, or longer movement durations, or both. ROIs in M1 and PMd arm zones returned time courses that were consistent with one another (Figure 5D, E). In both ROIs, the motion artifact was followed by an increase in positive reflectance, which is quite different from the time courses from the hand ROI (Figure 5C). Nevertheless, in the arm ROIs, negative reflectance increased from +3.5 from Cue and peaked at approximately the same time as the M1 hand ROI.

Two controls provided context for the time courses of the movement conditions (Figure 5C–E). First, in all ROIs, the withhold condition had time courses that remained close to baseline (Figure 5C–E). This control confirms that the reflectance change in the movement conditions was movement driven. Second, a ROI that comprised the entire forelimb representation of M1 and PMd (Figure 5B) returned time courses that differed entirely from the other three ROIs. Most importantly, the forelimb ROI lacked the characteristic negative peaks in Figure 5C–E. This control confirms that time courses from movement conditions were spatially specific responses that did not generalize to the entire forelimb representation.

Condition differences in cortical activity patterns

The spatial organization and size of the thresholded maps suggested similar cortical activity across conditions. Nevertheless, time courses (Figure 5C–E) and non-binarized average maps (Figure 6A) showed greater magnitudes of activity in the reach-to-grasp conditions as compared to the reach-only condition. This motivated us to directly compare cortical activity between conditions, which we achieved in two ways.

Figure 6

Download asset Open asset

First, we conducted t-test comparisons on movement frames (39-frame average) from condition pairs (Figure 6B, C). Our objective was to generate thresholded maps of pixels that darkened significantly in one movement condition relative to another movement condition. Maps thresholded at p < 0.0001 did not reveal distinguishable patches. After relaxing the threshold to p < 0.001, we found several patches in the M1 and PMd forelimb representations that darkened more in the reach-to-grasp conditions as compared to the reach-only condition (Figure 6B, C). In contrast, fewer and smaller patches reported differences between the precision and power conditions (not shown). Thus, the thresholded maps in Figure 6B, C flag cortical zones that were more active in the reach-to-grasp conditions as compared to the reach-only condition.

Second, we used cross-correlation at zero lag to compare average time series from pairs of conditions. We focused on the post-movement frames and correlated each frame from one condition with the time-matched frame in the paired condition. The analysis was limited to pixels within the forelimb representations. Correlation coefficients were pooled across monkeys such that each condition pair returned a distribution of 78 points (39 frames × 2 monkeys). A one-way analysis of variance (ANOVA) showed an effect of condition pairs on the correlation coefficients (F(2,231) = 12.98, p < 0.01; Figure 7D). Post hoc tests corrected for multiple comparison (n = 3; p < 0.01), showed that the correlation coefficients for precision X reach-only was lower than the other two condition pairs. The cross-correlations therefore show that the spatiotemporal patterns of cortical activity were most dissimilar between the precision and reach-only conditions. For context, cross-correlating movement conditions with the withhold condition returned a much lower distribution of coefficients (Figure 6D). Cross-correlation was therefore sensitive enough to detect condition differences in the time series. In sum, the two approaches that we used for directly comparing cortical activity between conditions indicate that condition differences were small but most pronounced between the precision and reach-only conditions.

Figure 7

Download asset Open asset

We investigated the relationship between movement-related cortical activity and motor maps in M1 and PMd. ISOI showed that cortical activity that supports reaching and grasping is concentrated in patches within the forelimb representations. The spatial configuration of the patches was consistent across task conditions despite variations in the reach-to-grasp movements. Nevertheless, the magnitude of activity within the patches scaled with forelimb use as measured from joint and muscle activity. The spatial organization of thresholded maps, and relative invariance across conditions, suggests that neural activity for arm and hand functions is not uniformly distributed throughout the forelimb representations. Instead, our observations are consistent with a functional organization wherein subzones of the forelimb representations are preferentially tuned for certain actions.

All data and code used in this paper are posted on OSF (https://osf.io/7sgbe/).

1. 1. Chehade N

   (2023) Open Science Framework

   Motor actions are spatially organized in motor and dorsal premotor cortex.

   https://doi.org/10.17605/OSF.IO/7SGBE

1. 1. Ances BM

   (2004) Coupling of changes in cerebral blood flow with neural activity: what must initially dip must come back up

   Journal of Cerebral Blood Flow and Metabolism 24:1–6.

   https://doi.org/10.1097/01.WCB.0000103920.96801.12

   • PubMed
   • Google Scholar
2. 1. Andersen P
   2. Hagan PJ
   3. Phillips CG
   4. Powell TP

   (1975) Mapping by microstimulation of overlapping projections from area 4 to motor units of the baboon’s hand

   Proceedings of the Royal Society of London. Series B, Biological Sciences 188:31–36.

   https://doi.org/10.1098/rspb.1975.0002

   • PubMed
   • Google Scholar
3. 1. Arieli A
   2. Grinvald A
   3. Slovin H

   (2002) Dural substitute for long-term imaging of cortical activity in behaving monkeys and its clinical implications

   Journal of Neuroscience Methods 114:119–133.

   https://doi.org/10.1016/s0165-0270(01)00507-6

   • PubMed
   • Google Scholar
4. 1. Baldwin MKL
   2. Cooke DF
   3. Krubitzer L

   (2017) Intracortical microstimulation maps of motor, somatosensory, and posterior parietal cortex in tree shrews (tupaia belangeri) reveal complex movement representations

   Cerebral Cortex 27:1439–1456.

   https://doi.org/10.1093/cercor/bhv329

   • PubMed
   • Google Scholar
5. 1. Bollimunta A
   2. Santacruz SR
   3. Eaton RW
   4. Xu PS
   5. Morrison JH
   6. Moxon KA
   7. Carmena JM
   8. Nassi JJ

   (2021) Head-mounted microendoscopic calcium imaging in dorsal premotor cortex of behaving rhesus macaque

   Cell Reports 35:109239.

   https://doi.org/10.1016/j.celrep.2021.109239

   • PubMed
   • Google Scholar
6. 1. Bonhoeffer T
   2. Grinvald A

   (1991) Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns

   Nature 353:429–431.

   https://doi.org/10.1038/353429a0

   • PubMed
   • Google Scholar
7. 1. Boudrias MH
   2. McPherson RL
   3. Frost SB
   4. Cheney PD

   (2010) Output properties and organization of the forelimb representation of motor areas on the lateral aspect of the hemisphere in rhesus macaques

   Cerebral Cortex 20:169–186.

   https://doi.org/10.1093/cercor/bhp084

   • PubMed
   • Google Scholar
8. 1. Card NS
   2. Gharbawie OA

   (2020) Principles of intrinsic motor cortex connectivity in primates

   The Journal of Neuroscience 40:4348–4362.

   https://doi.org/10.1523/JNEUROSCI.0003-20.2020

   • PubMed
   • Google Scholar
9. 1. Card NS
   2. Gharbawie OA

   (2022) Cortical connectivity is embedded in resting state at columnar resolution

   Progress in Neurobiology 213:102263.

   https://doi.org/10.1016/j.pneurobio.2022.102263

   • PubMed
   • Google Scholar
10. 1. Cavina-Pratesi C
    2. Monaco S
    3. Fattori P
    4. Galletti C
    5. McAdam TD
    6. Quinlan DJ
    7. Goodale MA
    8. Culham JC

    (2010) Functional magnetic resonance imaging reveals the neural substrates of arm transport and grip formation in reach-to-grasp actions in humans

    The Journal of Neuroscience 30:10306–10323.

    https://doi.org/10.1523/JNEUROSCI.2023-10.2010

    • PubMed
    • Google Scholar
11. 1. Chen LM
    2. Friedman RM
    3. Ramsden BM
    4. LaMotte RH
    5. Roe AW

    (2001) Fine-scale organization of SI (area 3b) in the squirrel monkey revealed with intrinsic optical imaging

    Journal of Neurophysiology 86:3011–3029.

    https://doi.org/10.1152/jn.2001.86.6.3011

    • PubMed
    • Google Scholar
12. 1. Chen-Bee CH
    2. Agoncillo T
    3. Xiong Y
    4. Frostig RD

    (2007) The triphasic intrinsic signal: implications for functional imaging

    The Journal of Neuroscience 27:4572–4586.

    https://doi.org/10.1523/JNEUROSCI.0326-07.2007

    • PubMed
    • Google Scholar
13. 1. Cooke DF
    2. Graziano MSA

    (2004) Super-flinchers and nerves of steel: defensive movements altered by chemical manipulation of a cortical motor area

    Neuron 43:585–593.

    https://doi.org/10.1016/j.neuron.2004.07.029

    • PubMed
    • Google Scholar
14. 1. Donoghue JP
    2. Leibovic S
    3. Sanes JN

    (1992) Organization of the forelimb area in squirrel monkey motor cortex: representation of digit, wrist, and elbow muscles

    Experimental Brain Research 89:1–19.

    https://doi.org/10.1007/BF00228996

    • PubMed
    • Google Scholar
15. 1. Ebina T
    2. Masamizu Y
    3. Tanaka YR
    4. Watakabe A
    5. Hirakawa R
    6. Hirayama Y
    7. Hira R
    8. Terada SI
    9. Koketsu D
    10. Hikosaka K
    11. Mizukami H
    12. Nambu A
    13. Sasaki E
    14. Yamamori T
    15. Matsuzaki M

    (2018) Two-photon imaging of neuronal activity in motor cortex of marmosets during upper-limb movement tasks

    Nature Communications 9:1879.

    https://doi.org/10.1038/s41467-018-04286-6

    • PubMed
    • Google Scholar
16. 1. Ejaz N
    2. Hamada M
    3. Diedrichsen J

    (2015) Hand use predicts the structure of representations in sensorimotor cortex

    Nature Neuroscience 18:1034–1040.

    https://doi.org/10.1038/nn.4038

    • PubMed
    • Google Scholar
17. 1. Fetz EE
    2. Cheney PD

    (1980) Postspike facilitation of forelimb muscle activity by primate corticomotoneuronal cells

    Journal of Neurophysiology 44:751–772.

    https://doi.org/10.1152/jn.1980.44.4.751

    • PubMed
    • Google Scholar
18. 1. Fiave PA
    2. Sharma S
    3. Jastorff J
    4. Nelissen K

    (2018) Investigating common coding of observed and executed actions in the monkey brain using cross-modal multi-variate fMRI classification

    NeuroImage 178:306–317.

    https://doi.org/10.1016/j.neuroimage.2018.05.043

    • PubMed
    • Google Scholar
19. 1. Friedman RM
    2. Chehade NG
    3. Roe AW
    4. Gharbawie OA

    (2020a) Optical imaging reveals functional domains in primate sensorimotor cortex

    NeuroImage 221:117188.

    https://doi.org/10.1016/j.neuroimage.2020.117188

    • PubMed
    • Google Scholar
20. 1. Friedman RM
    2. Morone KA
    3. Gharbawie OA
    4. Roe AW

    (2020b) Mapping mesoscale cortical connectivity in monkey sensorimotor cortex with optical imaging and microstimulation

    The Journal of Comparative Neurology 528:3095–3107.

    https://doi.org/10.1002/cne.24918

    • PubMed
    • Google Scholar
21. 1. Frostig RD
    2. Chen-Bee CH
    3. Johnson BA
    4. Jacobs NS

    (2017) Imaging Cajal’s neuronal avalanche: how wide-field optical imaging of the point-spread advanced the understanding of neocortical structure-function relationship

    Neurophotonics 4:031217.

    https://doi.org/10.1117/1.NPh.4.3.031217

    • PubMed
    • Google Scholar
22. 1. Gallivan JP
    2. McLean DA
    3. Valyear KF
    4. Pettypiece CE
    5. Culham JC

    (2011) Decoding action intentions from preparatory brain activity in human parieto-frontal networks

    The Journal of Neuroscience 31:9599–9610.

    https://doi.org/10.1523/JNEUROSCI.0080-11.2011

    • PubMed
    • Google Scholar
23. 1. Gallivan JP
    2. Culham JC

    (2015) Neural coding within human brain areas involved in actions

    Current Opinion in Neurobiology 33:141–149.

    https://doi.org/10.1016/j.conb.2015.03.012

    • PubMed
    • Google Scholar
24. 1. Gharbawie OA
    2. Stepniewska I
    3. Qi H
    4. Kaas JH

    (2011) Multiple parietal-frontal pathways mediate grasping in macaque monkeys

    The Journal of Neuroscience 31:11660–11677.

    https://doi.org/10.1523/JNEUROSCI.1777-11.2011

    • PubMed
    • Google Scholar
25. 1. Gould HJ
    2. Cusick CG
    3. Pons TP
    4. Kaas JH

    (1986) The relationship of corpus callosum connections to electrical stimulation maps of motor, supplementary motor, and the frontal eye fields in owl monkeys

    The Journal of Comparative Neurology 247:297–325.

    https://doi.org/10.1002/cne.902470303

    • PubMed
    • Google Scholar
26. 1. Graziano MSA
    2. Taylor CSR
    3. Moore T

    (2002) Complex movements evoked by microstimulation of precentral cortex

    Neuron 34:841–851.

    https://doi.org/10.1016/s0896-6273(02)00698-0

    • PubMed
    • Google Scholar
27. 1. Grèzes J
    2. Armony JL
    3. Rowe J
    4. Passingham RE

    (2003) Activations related to “mirror” and “canonical” neurones in the human brain: an fMRI study

    NeuroImage 18:928–937.

    https://doi.org/10.1016/s1053-8119(03)00042-9

    • PubMed
    • Google Scholar
28. 1. Grinvald A
    2. Lieke E
    3. Frostig RD
    4. Gilbert CD
    5. Wiesel TN

    (1986) Functional architecture of cortex revealed by optical imaging of intrinsic signals

    Nature 324:361–364.

    https://doi.org/10.1038/324361a0

    • PubMed
    • Google Scholar
29. 1. Grinvald A
    2. Frostig RD
    3. Siegel RM
    4. Bartfeld E

    (1991) High-resolution optical imaging of functional brain architecture in the awake monkey

    PNAS 88:11559–11563.

    https://doi.org/10.1073/pnas.88.24.11559

    • PubMed
    • Google Scholar
30. 1. Grinvald A
    2. Lieke EE
    3. Frostig RD
    4. Hildesheim R

    (1994) Cortical point-spread function and long-range lateral interactions revealed by real-time optical imaging of macaque monkey primary visual cortex

    The Journal of Neuroscience 14:2545–2568.

    https://doi.org/10.1523/JNEUROSCI.14-05-02545.1994

    • PubMed
    • Google Scholar
31. 1. Heider B
    2. Karnik A
    3. Ramalingam N
    4. Siegel RM

    (2010) Neural representation during visually guided reaching in macaque posterior parietal cortex

    Journal of Neurophysiology 104:3494–3509.

    https://doi.org/10.1152/jn.01050.2009

    • PubMed
    • Google Scholar
32. 1. Kaas JH
    2. Gharbawie OA
    3. Stepniewska I

    (2013) Cortical networks for ethologically relevant behaviors in primates

    American Journal of Primatology 75:407–414.

    https://doi.org/10.1002/ajp.22065

    • PubMed
    • Google Scholar
33. 1. Kim DS
    2. Duong TQ
    3. Kim SG

    (2000) High-resolution mapping of iso-orientation columns by fMRI

    Nature Neuroscience 3:164–169.

    https://doi.org/10.1038/72109

    • PubMed
    • Google Scholar
34. 1. Kondo T
    2. Saito R
    3. Otaka M
    4. Yoshino-Saito K
    5. Yamanaka A
    6. Yamamori T
    7. Watakabe A
    8. Mizukami H
    9. Schnitzer MJ
    10. Tanaka KF
    11. Ushiba J
    12. Okano H

    (2018) Calcium transient dynamics of neural ensembles in the primary motor cortex of naturally behaving monkeys

    Cell Reports 24:2191–2195.

    https://doi.org/10.1016/j.celrep.2018.07.057

    • PubMed
    • Google Scholar
35. Software

    1. Koon DJ

    (2008)

    B-Spline grid, image and point based registration, version Version 1.33.0.0

    MathWork.
36. 1. Lee S
    2. Wolberg G
    3. Shin SY

    (1997) Scattered data interpolation with multilevel B-splines

    IEEE Transactions on Visualization and Computer Graphics 3:228–244.

    https://doi.org/10.1109/2945.620490

    • Google Scholar
37. 1. Lemon RN
    2. Muir RB
    3. Mantel GW

    (1987) The effects upon the activity of hand and forearm muscles of intracortical stimulation in the vicinity of corticomotor neurones in the conscious monkey

    Experimental Brain Research 66:621–637.

    https://doi.org/10.1007/BF00270695

    • PubMed
    • Google Scholar
38. 1. Lu HD
    2. Roe AW

    (2007) Optical imaging of contrast response in Macaque monkey V1 and V2

    Cerebral Cortex 17:2675–2695.

    https://doi.org/10.1093/cercor/bhl177

    • PubMed
    • Google Scholar
39. 1. Malonek D
    2. Grinvald A

    (1996) InTeractions between electrical activity and cortical microcirculation revealed by imaging spectroscopy: Implications for functional brain mapping

    Science 272:551–554.

    https://doi.org/10.1126/science.272.5261.551

    • Google Scholar
40. 1. Menon RS
    2. Ogawa S
    3. Hu X
    4. Strupp JP
    5. Anderson P
    6. Uğurbil K

    (1995) BOLD based functional MRI at 4 Tesla includes a capillary bed contribution: echo-planar imaging correlates with previous optical imaging using intrinsic signals

    Magnetic Resonance in Medicine 33:453–459.

    https://doi.org/10.1002/mrm.1910330323

    • PubMed
    • Google Scholar
41. 1. Murphy JT
    2. Kwan HC
    3. MacKay WA
    4. Wong YC

    (1978) Spatial organization of precentral cortex in awake primates. III. Input-output coupling

    Journal of Neurophysiology 41:1132–1139.

    https://doi.org/10.1152/jn.1978.41.5.1132

    • PubMed
    • Google Scholar
42. 1. Nelissen K
    2. Vanduffel W

    (2011) Grasping-related functional magnetic resonance imaging brain responses in the macaque monkey

    The Journal of Neuroscience 31:8220–8229.

    https://doi.org/10.1523/JNEUROSCI.0623-11.2011

    • PubMed
    • Google Scholar
43. 1. Nelissen K
    2. Fiave PA
    3. Vanduffel W

    (2018) Decoding grasping movements from the parieto-frontal reaching circuit in the nonhuman primate

    Cerebral Cortex 28:1245–1259.

    https://doi.org/10.1093/cercor/bhx037

    • PubMed
    • Google Scholar
44. 1. Papadourakis V
    2. Raos V

    (2019) Neurons in the macaque dorsal premotor cortex respond to execution and observation of actions

    Cerebral Cortex 29:4223–4237.

    https://doi.org/10.1093/cercor/bhy304

    • PubMed
    • Google Scholar
45. 1. Park MC
    2. Belhaj-Saïf A
    3. Gordon M
    4. Cheney PD

    (2001) Consistent features in the forelimb representation of primary motor cortex in rhesus macaques

    The Journal of Neuroscience 21:2784–2792.

    https://doi.org/10.1523/JNEUROSCI.21-08-02784.2001

    • PubMed
    • Google Scholar
46. 1. Peters AJ
    2. Chen SX
    3. Komiyama T

    (2014) Emergence of reproducible spatiotemporal activity during motor learning

    Nature 510:263–267.

    https://doi.org/10.1038/nature13235

    • PubMed
    • Google Scholar
47. 1. Picard N
    2. Matsuzaka Y
    3. Strick PL

    (2013) Extended practice of a motor skill is associated with reduced metabolic activity in M1

    Nature Neuroscience 16:1340–1347.

    https://doi.org/10.1038/nn.3477

    • PubMed
    • Google Scholar
48. 1. Raos V
    2. Franchi G
    3. Gallese V
    4. Fogassi L

    (2003) Somatotopic organization of the lateral part of area F2 (dorsal premotor cortex) of the macaque monkey

    Journal of Neurophysiology 89:1503–1518.

    https://doi.org/10.1152/jn.00661.2002

    • PubMed
    • Google Scholar
49. 1. Raos V
    2. Evangeliou MN
    3. Savaki HE

    (2007) Mental simulation of action in the service of action perception

    The Journal of Neuroscience 27:12675–12683.

    https://doi.org/10.1523/JNEUROSCI.2988-07.2007

    • PubMed
    • Google Scholar
50. 1. Rathelot JA
    2. Strick PL

    (2006) Muscle representation in the macaque motor cortex: an anatomical perspective

    PNAS 103:8257–8262.

    https://doi.org/10.1073/pnas.0602933103

    • PubMed
    • Google Scholar
51. 1. Rouse AG
    2. Schieber MH

    (2016) Spatiotemporal distribution of location and object effects in primary motor cortex neurons during reach-to-grasp

    The Journal of Neuroscience 36:10640–10653.

    https://doi.org/10.1523/JNEUROSCI.1716-16.2016

    • PubMed
    • Google Scholar
52. 1. Rueckert D
    2. Sonoda LI
    3. Hayes C
    4. Hill DLG
    5. Leach MO
    6. Hawkes DJ

    (1999) Nonrigid registration using free-form deformations: application to breast MR images

    IEEE Transactions on Medical Imaging 18:712–721.

    https://doi.org/10.1109/42.796284

    • PubMed
    • Google Scholar
53. 1. Ruiz O
    2. Lustig BR
    3. Nassi JJ
    4. Cetin A
    5. Reynolds JH
    6. Albright TD
    7. Callaway EM
    8. Stoner GR
    9. Roe AW

    (2013) Optogenetics through windows on the brain in the nonhuman primate

    Journal of Neurophysiology 110:1455–1467.

    https://doi.org/10.1152/jn.00153.2013

    • PubMed
    • Google Scholar
54. 1. Saleh M
    2. Takahashi K
    3. Hatsopoulos NG

    (2012) Encoding of coordinated reach and grasp trajectories in primary motor cortex

    The Journal of Neuroscience 32:1220–1232.

    https://doi.org/10.1523/JNEUROSCI.2438-11.2012

    • PubMed
    • Google Scholar
55. 1. Sessle BJ
    2. Wiesendanger M

    (1982) Structural and functional definition of the motor cortex in the monkey (Macaca fascicularis)

    The Journal of Physiology 323:245–265.

    https://doi.org/10.1113/jphysiol.1982.sp014071

    • PubMed
    • Google Scholar
56. 1. Shtoyerman E
    2. Arieli A
    3. Slovin H
    4. Vanzetta I
    5. Grinvald A

    (2000) Long-term optical imaging and spectroscopy reveal mechanisms underlying the intrinsic signal and stability of cortical maps in V1 of behaving monkeys

    The Journal of Neuroscience 20:8111–8121.

    https://doi.org/10.1523/JNEUROSCI.20-21-08111.2000

    • PubMed
    • Google Scholar
57. 1. Siegel RM
    2. Duann JR
    3. Jung TP
    4. Sejnowski T

    (2007) Spatiotemporal dynamics of the functional architecture for gain fields in inferior parietal lobule of behaving monkey

    Cerebral Cortex 17:378–390.

    https://doi.org/10.1093/cercor/bhj155

    • PubMed
    • Google Scholar
58. 1. Sirotin YB
    2. Das A

    (2009) Anticipatory haemodynamic signals in sensory cortex not predicted by local neuronal activity

    Nature 457:475–479.

    https://doi.org/10.1038/nature07664

    • PubMed
    • Google Scholar
59. 1. Sirotin YB
    2. Hillman EMC
    3. Bordier C
    4. Das A

    (2009) Spatiotemporal precision and hemodynamic mechanism of optical point spreads in alert primates

    PNAS 106:18390–18395.

    https://doi.org/10.1073/pnas.0905509106

    • PubMed
    • Google Scholar
60. 1. Stepniewska I
    2. Fang PC
    3. Kaas JH

    (2005) Microstimulation reveals specialized subregions for different complex movements in posterior parietal cortex of prosimian galagos

    PNAS 102:4878–4883.

    https://doi.org/10.1073/pnas.0501048102

    • PubMed
    • Google Scholar
61. 1. Tanigawa H
    2. Lu HD
    3. Roe AW

    (2010) Functional organization for color and orientation in macaque V4

    Nature Neuroscience 13:1542–1548.

    https://doi.org/10.1038/nn.2676

    • PubMed
    • Google Scholar
62. 1. Trautmann EM
    2. O’Shea DJ
    3. Sun X
    4. Marshel JH
    5. Crow A
    6. Hsueh B
    7. Vesuna S
    8. Cofer L
    9. Bohner G
    10. Allen W
    11. Kauvar I
    12. Quirin S
    13. MacDougall M
    14. Chen Y
    15. Whitmire MP
    16. Ramakrishnan C
    17. Sahani M
    18. Seidemann E
    19. Ryu SI
    20. Deisseroth K
    21. Shenoy KV

    (2021) Dendritic calcium signals in rhesus macaque motor cortex drive an optical brain-computer interface

    Nature Communications 12:3689.

    https://doi.org/10.1038/s41467-021-23884-5

    • PubMed
    • Google Scholar
63. 1. Ts’o DY
    2. Frostig RD
    3. Lieke EE
    4. Grinvald A

    (1990) Functional organization of primate visual cortex revealed by high resolution optical imaging

    Science 249:417–420.

    https://doi.org/10.1126/science.2165630

    • PubMed
    • Google Scholar
64. 1. Vanzetta I
    2. Slovin H
    3. Omer DB
    4. Grinvald A

    (2004) Columnar resolution of blood volume and oximetry functional maps in the behaving monkey; implications for FMRI

    Neuron 42:843–854.

    https://doi.org/10.1016/j.neuron.2004.04.004

    • PubMed
    • Google Scholar
65. 1. Vargas-Irwin CE
    2. Shakhnarovich G
    3. Yadollahpour P
    4. Mislow JMK
    5. Black MJ
    6. Donoghue JP

    (2010) Decoding complete reach and grasp actions from local primary motor cortex populations

    The Journal of Neuroscience 30:9659–9669.

    https://doi.org/10.1523/JNEUROSCI.5443-09.2010

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Neural activity is spatially clustered in motor and dorsal premotor cortex" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Joshua Gold as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Wim Vanduffel (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) The evidence for "clustered" activity is incomplete. It is not clear whether there is convincing evidence that only a small part of the arm/hand representation is activated. Showing that the non-active regions can be activated by other actions would have been the most direct and convincing evidence here (see main comment 1. from both reviewers 1 and 2).

2) Both reviewers also point out that the comparison between activation and stimulation data (which is the unique feature of this dataset) is rather superficial and both reviewers saw a lot of potential here.

Reviewer #1 (Recommendations for the authors):

This paper describes the neural activity, measured by intrinsic optical imaging in reach-to-grasp, and reach-only conditions in relation to the Intra-cortical micro stimulation maps. The paper mostly describes a relatively unique and potentially useful data set. However, in the current version, no real hypotheses about the organization of M1 and PMd are tested convincingly. For example, the claim of "clustered neural activity" is not tested against any quantifiable alternative hypothesis of non-clustered activity, and support for this idea is therefore incomplete.

The combination of intrinsic optical imaging and intra-cortical micro-stimulation of the motor system of two macaque monkeys promised to be a unique and highly interesting dataset. The experiments are carefully conducted. In the analysis and interpretation of the results, however, the paper was disappointing to me. The two main weaknesses in my mind were:

1. The alternative hypotheses depicted in Figure 1B are not subjected to any quantifiable test. When is an activity considered to be clustered and when is it distributed? The fact that the observed actions only activate a small portion of the forelimb area (Figure 5G, H) is utterly unconvincing, as this analysis is highly threshold-dependent. Furthermore, it could be the case that the non-activated regions simply do not give a good intrinsic signal, as they are close to microvasculature (something that you actually seem to argue in Figure 6b). Until the authors can show that the other parts of the forelimb area are clearly activated for other forelimb actions (as you suggest on line 625), I believe the claim of cluster neural activity stands unsupported.

2. The most interesting part of the study (which cannot be easily replicated with human fMRI studies) is the correspondence between the evoked activity and intra-cortical stimulation maps. However, this is impeded by the subjective and low-dimensional description of the evoked movement during stimulation (mainly classifying the moving body part), and the relatively low-dimensional nature (4 conditions) of the evoked activity.

3. Many details about the statistical analysis remain unclear and seem not well motivated.

Figure 1B: What does FL (gray area, caption) stand for? I can't see that the term is defined.

Line 158: I gather that unsuccessful trials were repeated. What were the repetition rates in the three conditions? Was the big and small sphere (here listed as one condition) also varied in a pseudorandom manner?

Line 223: I gather from this that the movement elicited in the ICMS was judged qualitatively by the two experimenters, but no EMG / movement kinematics was recorded as in the overt behavior. Did you make any attempts to validate the experimenter's judgement against objective recordings? Could you calculate inter-rater reliability or did the two experimenters interact with each other during the judgement?

Line 260: The baseline was either the first frame or the initial first four frames. How did you decide when to use which method? The statement "to increase signal-to-noise ratio" is somewhat cryptic. Increase in respect to what? Baseline subtraction clearly needs to be done in some form or the other. If you wanted a stable baseline, why not take the entire period before cue appearance?

Line 264-266. When you state: kernel = 250 or 550 pixels and low pass filer 5 or 15 pixels does this refer to the data from the two different monkeys? How were these values decided?

Line 278: If you do another baseline subtraction here, is the baseline subtraction on line 260 not superfluous? When did you use 5 and when 9 frames?

For the fMRI crowd, could you add here how pixel darkening and brightening relate to blood flow changes and blood oxygenation changes?

Line 305: Your nomenclature of cells / ROIs is somewhat confusing. What was the motivation to spatially average pixels across each cell/ROI instead of entering all pixels into the clustering analysis? What do mean by "Spatially matched cells"? Was co-registering not something that was done on the entire window?

Line 312: How big was the grid in horizontal and vertical dimensions for each monkey? Is this where the 1700 / 2249 numbers come from?

Line 313: A distances metric is defined to be zeros to itself, dist(A,A) = 0. The correlation of A with itself is 1. Do you mean to say that you used 1-max_correlation, or that you used the correlation as a similarity metric?

Line 323: My guess the point of this analysis is that the total distances decrease with an increasing number of clusters (and the correlation increases with an increasing number of clusters). So, fitting a line assumes a linear relationship between these two variables. When you say "The longest of those orthogonal lines identified the optimal number of clusters" why are the lines orthogonal? By longest – do you mean the largest deviation between the linear fit either above or below the line? This would pick either an especially good or bad parcellation. Or did you look at deviations above or below only?

Line 327: In the methods, we have 3 conditions – in the results 4. Which one is correct?

Line 370: Is the darkening appearing at 1s an artifact of the measurement or real? At least for the posterior border, it looks like it follows exactly the border of the window.

Line 451: "that shifted and expanded across the time series" – is there a statistical test for the claim that the activity region shifted?

Line 462: To what degree is the fact the activity averaged over the entire window remains stable simply a consequence of the high-pass filtering applied to each image? High-pass filtering basically removes the mean across the entire window, no?

Line 466: What is meant by "domains"? Activity clusters?

Line 5E: The analysis here involves statistical testing within each session, and then averaging activity estimates across sessions, subject to an arbitrary 50% threshold for statistical significance within the session. I cannot see a good motivation for this awkward type of analysis. If you want to consider trials a random effect, I would recommend calculating a statistical test for all trials across sessions combined. If you want to consider sessions as a random effect, then calculate a t-test across the 8 repeated activity estimates per session – this analysis also automatically takes the variability across trials into account. Or is there another reason why you chose this baroque style of analysis that is mentioned?

Line 474: "organization of domains was more similar across conditions within an animal than for the same condition across animals". See Ejaz et al. (2015). Nature Neuroscience, for a similar observation in human fMRI motor maps.

Line 483: The overlap analysis is unfortunately quite dependent on the threshold.

Line 494: form → from?

I am not sure what I am supposed to learn from the cluster analysis. It is entitled "Clustering time courses recapitulates spatial patterns of activity" – likely referring to the similarity between Figure 6a and 6b. As the clustering partly depends on the magnitude of the signal at the time points presented in Figure 6a, what is the alternative? Is this similarity more than expected by chance?

Line 527-546. The overlap between the maps is very hard to interpret, as it is highly dependent on the threshold procedure applied, which is somewhat arbitrarily chosen (see above). Even random maps of a certain size would overlap to some degree, and even identical "true" maps don't overlap completely due to measurement error. An overlap of 0.5 has no intrinsic meaning. It would probably be more informative to report the correlation between unthresholded maps, and compare this to a noise ceiling, where you correlate the maps of one condition across sessions.

Reviewer #2 (Recommendations for the authors):

Chehade and Gharbawie investigated motor and premotor cortex in macaque monkeys performing grasping and reaching tasks. They used intrinsic signal optical imaging (ISOI) covering an exceedingly large field-of-view extending from the IPS to the PS. They compared reaching and fine/power-grip grasping ISOI maps with "motor" maps which they obtained using extensive intracranial microstimulation. The grasping/reaching-induced activity activated relatively isolated portions of M1 and PMd, and did not cover the entire ICM-induced 'motor' maps of the upper limbs. The authors suggest that small subzones exist in M1 and PMd that are preferentially activated by different types of forelimb actions. In general, the authors address an important topic. The results are not only highly relevant for increasing our basic understanding of the functional architecture of the motor-premotor cortex and how it represents different types of forelimb actions, but also for the development of brain-machine interfaces. These are challenging experiments to perform and add to the existing yet complementary electrophysiology, fMRI, and optical imaging experiments that have been performed on this topic – due to the high sensitivity and large coverage of the particular IOSI methods employed by the authors. The manuscript is generally well written and the analyses seem overall adequate – but see below for some additional analyses that should be done. Although I'm generally enthusiastic about this manuscript, there are two major issues that should be clarified. These major questions relate mainly to potential thresholding issues and clustering issues.

1) The main claim of the authors is that specific forelimb actions activate only a small fraction of what they call the motor map (i.e., those parts of M1/PMd that evoke muscle contractions upon ICM). The action-related activity is measured by ISOI. When looking a the 'raw' reflectance maps, it is rather clear that relatively wide portions of the exposed cortex are activated by grasping/reaching, especially at later time points after the action. In fact, another reading of the results may be that there are two zones of 'deactivation' that split a large swath of motor-premotor cortex being activated by the grasping/reaching actions. (e.g. at 6 seconds after the cue in Figure 3A, 5A). At first sight, the 'deactivated' regions seem to be located in the cortex representing the trunk/shoulder/face – hence regions not necessarily activated (or only weakly) during the grasping/reaching actions. If true, this means that most of the relevant M1/PMd cortex IS activated during the latter actions – opposing the 'clustering' claims of the authors. This raises the question of whether the 'granularity' claimed by the authors is:

1. Threshold dependent. In this context, the authors should provide an analysis whereby 'granularity' is shown independent of statistical thresholds of the ISOI maps.

2. Dependent on the time-point one assesses the maps. Given the sluggish hemodynamic responses, it is unclear which part of the ISOI maps conveys the most information relative to the cue and arm/hand movements. I suspect that timepoints > 6 s will reveal even larger 'homogeneous' activations compared to the maps < 6s.

In fact, Figure 5F (which is highly thresholded) shows a surprisingly good match between the different forelimb actions, which argues against the existence of small subzones that are preferentially activated by different types of forelimb actions -the main claim of the authors.

2) Related to the previous point, the ROI selections/definitions for the time course analyses seem highly arbitrary. As indicated in the introduction, the clustering hypothesis dictates that "an arm function would be concentrated in subzones of the motor arm zones. Neural activity in adjacent subzones would be tuned for other arm functions." To test this hypothesis directly in a straightforward manner, the authors could use the results from the ICM experiment to construct independent ROIs and to evaluate the ISOI responses for the different actions. In that case, the authors could do a straightforward ANOVA (if the data permits parametric analyses) with ROI, action, and time point (and possibly subject) as factors.

3) More details about the transparent silicone membrane should be provided: How implanted? How maintained? Please provide pictures at the end of the 9 and 22months periods.

4) Figure 1D: Are the yellow circles task cues that the monkeys saw? If not, what were the different cues (Reach, Grasp, Withhold)? Were precision and power grip (and reach?) trials similarly cued?

5) For the MEG activity a total of 14 27gauge wires were inserted in 7 muscles. This sounds rather invasive. Did monkeys tolerate this easily? Were local anesthetics used? How deep were the wires inserted in the muscle? How did one prevent the monkeys grabbing these wires?

6) The large FOV was illuminated using 2-3 LEDs. Was illumination uniform throughout the FOV? If not, can this lead to inhomogeneous sensitivity?

7) Pixel values were clipped, if I understood correctly, at either 0.3 SD or 1 SD from the mean. This significantly changes the dynamic range of these pixel values. Are results different without clipping?

8) An arbitrarily p<0.0001 statistical threshold was used. Why was no correction for multiple correspondence performed as one collects data from ~ 600k-1400k pixels (yet without considering the masks)?

9) What is the reason for obvious edge effects in the withhold condition (Figure 3A, 3F) and apparently also the precision grip condition (Figure 5A: 1.1-3.5sec)? Is this artifactual?

10) How are the ROIs defined (cfr major point 2)? They differ in size and location between analyses (e.g. Figure 3 vs Figure 5/S1). Please discuss.

11) I find it a bit counter-intuitive that the same color-code (i.e. black values) is given for 'activations' in panel 3G and 'deactivations' in panel 3H. Why not using red and blue instead? Along the same vein: activations in the time-courses are negative (reflecting the darkening). It would be instructive for the non-experienced reader to either add activation/ suppression on the figures (above and below zero in Figures 3I and J and 5C), or to invert the Y-axis.

12) Line 426: Figure 5K does not exist.

13) Line 520: It is argued that the unsupervised cluster analysis, which is quite interesting, is similar in both monkeys. However, this is not obvious from the data: neither in the spatial domain (Figure 6B vs S2B) nor in the temporal domain Figure 6C and S2 C, especially the blue plots. In fact, the clustering data from monkey G reveal a rather widespread, uninterrupted pattern arguing against the 'cluster hypothesis' of the authors. This should be discussed in more depth.

14) Figure 7A: It is unclear which analysis was done. Is this simply giving 3 different colors to each pixel – indicating 1) precision > baseline; 2) reach > baseline 3) both > baseline (for the left panel)? Why not performing straight (pair-wise) subtractions?

Essential revisions:

1) The evidence for "clustered" activity is incomplete. It is not clear whether there is convincing evidence that only a small part of the arm/hand representation is activated. Showing that the non-active regions can be activated by other actions would have been the most direct and convincing evidence here (see main comment 1. from both reviewers 1 and 2).

We appreciate the reviewers’ concerns. We overhauled that analyses for measuring the spatial extent of cortical activity. We present new and more robust evidence that cortical activity overlapped relatively small portions (<40%) of M1 and PMd forelimb representations (Figure 5 and Figure 5 – supplemental Figure 1-2).

We agree with the reviewers; had the non-active regions shown activation in other arm/hand actions, that would have supported the notion of functionally specific subregions. Obtaining that data would require optical imaging while monkeys perform a variety of tasks that test a range of arm and hand movements (e.g., wide range of finger movements, wrist postures, arm directions, etc.). The two monkeys that supplied data for this study were trained only on a reach-to-grasp task with few task conditions. Moreover, the monkeys are no longer available for optical imaging. Thus, obtaining the desired data would involve conducting a larger version of the present study in naïve monkeys. A conservative estimate for completing such a study is 2-3 years. Until we can complete such a study, we expanded the Discussion to consider limitations of the behavioral task and our interpretation of the results.

2) Both reviewers also point out that the comparison between activation and stimulation data (which is the unique feature of this dataset) is rather superficial and both reviewers saw a lot of potential here.

We agree with the reviewers’ concerns. We now leverage each monkey’s motor map to guide placement of regions-of-interest (ROIs) for measuring time courses. Similarly, we use the motor map to spatially constrain the field-of-view for analyses of the activity maps.

Reviewer #1 (Recommendations for the authors):

This paper describes the neural activity, measured by intrinsic optical imaging in reach-to-grasp, and reach-only conditions in relation to the Intra-cortical micro stimulation maps. The paper mostly describes a relatively unique and potentially useful data set. However, in the current version, no real hypotheses about the organization of M1 and PMd are tested convincingly. For example, the claim of "clustered neural activity" is not tested against any quantifiable alternative hypothesis of non-clustered activity, and support for this idea is therefore incomplete.

The combination of intrinsic optical imaging and intra-cortical micro-stimulation of the motor system of two macaque monkeys promised to be a unique and highly interesting dataset. The experiments are carefully conducted. In the analysis and interpretation of the results, however, the paper was disappointing to me. The two main weaknesses in my mind were:

a) The alternative hypotheses depicted in Figure 1B are not subjected to any quantifiable test. When is an activity considered to be clustered and when is it distributed? The fact that the observed actions only activate a small portion of the forelimb area (Figure 5G, H) is utterly unconvincing, as this analysis is highly threshold-dependent. Furthermore, it could be the case that the non-activated regions simply do not give a good intrinsic signal, as they are close to microvasculature (something that you actually seem to argue in Figure 6b). Until the authors can show that the other parts of the forelimb area are clearly activated for other forelimb actions (as you suggest on line 625), I believe the claim of cluster neural activity stands unsupported.

We appreciate the reviewer’s concerns and we have made several revisions.

(1) The two panels in Figure 1B should have been presented as potential outcomes as opposed to hypotheses in need of quantifiable testing. We revised the Introduction (line 105-111) and the Results (line 149-152) accordingly.

(2) We agree that the thresholding procedure adopted in the original submission could have impacted the spatial measurements of cortical activity (i.e., Figure 5G-H in original submission). We have completely revised the thresholding procedure and it is now based on statistical comparisons that include all trials (instead of thresholding by number of sessions in the original submission). Thus, the thresholded maps in Figure 5G & 5J are now obtained from pixel-by-pixel comparisons (t-tests, p<1e-4) between frames acquired post-movement and frames acquired before movement. Nevertheless, even with this relatively relaxed threshold, the largest activity maps overlapped <40% of the forelimb representations.

It is important to note that major vessels were excluded from the thresholded map and from the motor map. Thus, uncertainty about imaging in and around vessels was likely not a factor in the calculated overlap between thresholded maps and the motor map.

(3) We agree that showing activation in other parts of the forelimb representations in response to action other than reach-to-grasp would have supported some of the arguments that we previously put forth. Unfortunately, we do not have the supporting data and obtaining it would take months/years. We have therefore expanded the Discussion to include limitations of the behavioral task (line 439-443).

b) The most interesting part of the study (which cannot be easily replicated with human fMRI studies) is the correspondence between the evoked activity and intra-cortical stimulation maps. However, this is impeded by the subjective and low-dimensional description of the evoked movement during stimulation (mainly classifying the moving body part), and the relatively low-dimensional nature (4 conditions) of the evoked activity.

We agree with the reviewer on all accounts. We expanded the Discussion to consider the low dimensionality of the motor maps and the behavioral task (line 439-449).

Measuring cortical activity in a variety of motor tasks would likely have provided additional insight about movement-related cortical activity. Nevertheless, including additional tasks, even if it were possible to do so in the same monkeys, would have delayed study completion by months/years. The hidden challenge of the experimental design is that each monkey is trained to not move for many seconds to minimize contamination of ISOI signals. For example, from trial initiation to Go Cue, the monkey must hold its hand in the start position for 5 seconds. Similarly, after movement completion, the monkey must hold its hand in the start position for another 5 seconds. In between successful trials, a monkey must wait for ~12 seconds before it can initiate a new trial. These durations are >1 order of magnitude longer than in electrophysiological studies in comparable tasks. Achieving consistent task performance with the long durations used here, took months of daily training. Moreover, our monkeys typically run out of steam after ~60-70 min of working on the task. This forces us to limit the overall number of task conditions tested in a session, to obtain a large enough number of trials from each condition.

c) Many details about the statistical analysis remain unclear and seem not well motivated.

We address the reviewer’s specific concerns below.

Figure 1B: What does FL (gray area, caption) stand for? I can't see that the term is defined.

FL is short for forelimb. The term is now defined in the figure caption.

Line 158: I gather that unsuccessful trials were repeated. What were the repetition rates in the three conditions? Was the big and small sphere (here listed as one condition) also varied in a pseudorandom manner?

All conditions, including the large and small spheres, were presented in pseudorandom order (line 624-625). We revised the Methods to clarify that the large and small sphere are considered separate conditions (line 593-594).

We calculated the average number of repetitions for each condition. Results are pooled across monkeys.

• Median (IQR)

• Reach-to-grasp with precision grip: 0.6 (0.2 – 1.3)

• Reach-to-grasp with power grip: 0.4 (0.2 – 0.7)

• Reach only: 0.3 (0.1 – 0.7)

• Withhold: 0.3 (0.1 – 0.4)

We also calculated the average failure rate for each condition and report those numbers in the Methods (line 607-609; 615; 622-623). Results are pooled across monkeys.

• Median (IQR)

• Reach-to-grasp with precision grip: 18%; (13 – 39%)

• Reach-to-grasp with power grip: 16%; (9 – 28%)

• Reach only: 12%; (4 – 32%)

• Withhold: 12%; (7 – 19%)

Line 223: I gather from this that the movement elicited in the ICMS was judged qualitatively by the two experimenters, but no EMG / movement kinematics was recorded as in the overt behavior. Did you make any attempts to validate the experimenter's judgement against objective recordings? Could you calculate inter-rater reliability or did the two experimenters interact with each other during the judgement?

In previous experiments in macaques and squirrel monkeys, we benchmarked experimenter classification of ICMS-evoked movement against ICMS-evoked EMG. In winner-take-all classification like the one adopted here; experimenter classification of evoked movements were consistent with the largest EMG response. Nevertheless, that benchmarking was not redone for this study. Also, inter-rate reliability cannot be assessed because the experimenters collaborated on movement classification. We expanded the Methods (line 708-711) and Discussion to reflect these limitations (line 443-449).

Line 260: The baseline was either the first frame or the initial first four frames. How did you decide when to use which method? The statement "to increase signal-to-noise ratio" is somewhat cryptic. Increase in respect to what? Baseline subtraction clearly needs to be done in some form or the other. If you wanted a stable baseline, why not take the entire period before cue appearance?

We agree with the reviewer on all points. First-frame subtraction has now been standardized across monkeys and imaging sessions. The first frame is now an average of 10 frames acquired from -1.0 to 0 s from Cue. We revised all analyses and figures accordingly and noted little/no effect on the results. Also, for clarity, we revised the statement about signal-to-noise ratio (line 745-748).

Line 264-266. When you state: kernel = 250 or 550 pixels and low pass filer 5 or 15 pixels does this refer to the data from the two different monkeys? How were these values decided?

Yes, in the original submission, kernel sizes differed between monkeys. We have reprocessed the data so that filtering parameters are identical for both monkeys. High pass kernel = 250 pixels. Low pass kernel = 5 pixels (line 748-750).

Line 278: If you do another baseline subtraction here, is the baseline subtraction on line 260 not superfluous? When did you use 5 and when 9 frames?

We revised the first-frame subtraction (i.e., baseline subtraction) to ensure that it is done consistently across sessions and monkeys (line 745-748).

We agree that a second baseline subtraction would make the first one superfluous. But only one subtraction was done: the first-frame was subtracted from all subsequent frames in the same trial (line 260 in original submission). After first-frame subtraction, we average baseline frames and separately average movement frames, to statistically compare them to each other.

For the fMRI crowd, could you add here how pixel darkening and brightening relate to blood flow changes and blood oxygenation changes?

We expanded this section to describe pixel darkening and brightening and to relate it to hemodynamic signals recorded in fMRI (line 769-778). In brief, for the present illumination (630 nm wavelength), pixel darkening is established as an indicator of increased deoxyhemoglobin (HbR) concentrations. Pixel darkening in ISOI has been likened to the “initial dip” in BOLD fMRI, which has been elusive in some fMRI studies. In contrast pixel brightening in ISOI is considered an indicator of increased concentrations of oxygenated hemoglobin (HbO) and blood volume/flow. ISOI pixel brightening has therefore been likened to an increase in BOLD fMRI.

Line 305: Your nomenclature of cells / ROIs is somewhat confusing. What was the motivation to spatially average pixels across each cell/ROI instead of entering all pixels into the clustering analysis? What do mean by "Spatially matched cells"? Was co-registering not something that was done on the entire window?

W averaged pixels within a cell for: (1) denoising, and (2) speeding up computation. Yes, co-registration was done for the entire field-of-view. We should have referred to that concept instead of the confusing “spatially matched cells”.

We agree with the reviewer that aspects of the clustering analysis, including nomenclature, were confusing. We redid the clustering analysis and believe that the revisions would have addressed the points raised here and elsewhere by both reviewers. Nevertheless, we decided to replace the clustering analysis with a more straightforward analysis that is presented (line 345-374 and Figure 6).

The purpose of clustering was to extract spatial patterns of activity from time series. We were enthusiastic about clustering because of its potential to return meaningful activity maps without input from us about frame range. To partially preserve that feature in the new analysis, we used all frames acquired post movement (i.e., 39 frames captured from +2.2 to +6.0s from Cue). We chose this frame range to (1) exclude frames that may contain motion artifact, and (2) focus on frames likely to have task-related activity.

Line 312: How big was the grid in horizontal and vertical dimensions for each monkey? Is this where the 1700 / 2249 numbers come from?

This inquiry pertains to the clustering analysis, which has been removed from the manuscript.

Nevertheless, yes, the numbers refer to the dimensions of the grid. The grids in both monkeys had identical size cells. But the number of cells differed between monkeys (1700 vs 2249) because of small differences in chamber placement and cortical anatomy.

Line 313: A distances metric is defined to be zeros to itself, dist(A,A) = 0. The correlation of A with itself is 1. Do you mean to say that you used 1-max_correlation, or that you used the correlation as a similarity metric?

This inquiry pertains to the clustering analysis, which has been removed from the manuscript. Nevertheless, to address the reviewer’s question, correlation was indeed used as a similarity metric.

Line 323: My guess the point of this analysis is that the total distances decrease with an increasing number of clusters (and the correlation increases with an increasing number of clusters). So, fitting a line assumes a linear relationship between these two variables. When you say "The longest of those orthogonal lines identified the optimal number of clusters" why are the lines orthogonal? By longest – do you mean the largest deviation between the linear fit either above or below the line? This would pick either an especially good or bad parcellation. Or did you look at deviations above or below only?

This inquiry pertains to the clustering analysis, which has been removed from the manuscript. Nevertheless, to address the reviewer’s question, the orthogonal lines were always below the fitted line. This is a function of how the line was fitted to the graph in the first place. Thus, criteria were consistent for selecting clusters.

Line 327: In the methods, we have 3 conditions – in the results 4. Which one is correct?

The task involved 4 total conditions: 3 movement conditions and 1 withhold condition. We have checked the entire manuscript and revised any inconsistencies.

Line 370: Is the darkening appearing at 1s an artifact of the measurement or real? At least for the posterior border, it looks like it follows exactly the border of the window.

The posterior border of the field-of-view coincides with the central sulcus, which contains a major blood vessel. The darkening is most likely due to reflectance change in vessels as opposed to an edge effect of the cranial window. We revised the text to indicate this point (line 157-162)

Line 451: "that shifted and expanded across the time series" – is there a statistical test for the claim that the activity region shifted?

We revised the statement to convey that it is a qualitative description of the spatio-temporal pattern of cortical activity (line 254-255).

Line 462: To what degree is the fact the activity averaged over the entire window remains stable simply a consequence of the high-pass filtering applied to each image? High-pass filtering basically removes the mean across the entire window, no?

We tested the impact of the high-pass filter on the reflectance change time courses from a region-of-interest that covers the entire field of view. Thus, we systematically varied the size of the filtering kernel (100-800 pixels in steps of 100 pixels) then examined the time courses. The results of those tests are plotted in Author response image 1 for the precision grip (movement condition) and withhold conditions (no movement). We note that in both conditions, most of the colored lines overlapped. This indicates that kernel size had little/no impact on time courses measured from the large ROIs. Nevertheless, the necessity for high pass filtering is evident from the gray time course, which was generated without high pass filtering. That line plot shows the dominance of global signals over the intrinsic signals of interest.

Author response image 1

Download asset Open asset

Line 466: What is meant by "domains"? Activity clusters?

Yes, domains are activity clusters. The term was defined on line 357 in the original submission (now line 145-149).

Line 5E: The analysis here involves statistical testing within each session, and then averaging activity estimates across sessions, subject to an arbitrary 50% threshold for statistical significance within the session. I cannot see a good motivation for this awkward type of analysis. If you want to consider trials a random effect, I would recommend calculating a statistical test for all trials across sessions combined. If you want to consider sessions as a random effect, then calculate a t-test across the 8 repeated activity estimates per session – this analysis also automatically takes the variability across trials into account. Or is there another reason why you chose this baroque style of analysis that is mentioned?

We agree with the reviewer and have overhauled the analysis. We now pool trials across sessions then conduct a t-test. Thus, for every condition, only one t-test is calculated (Figure 5G and 5J).

Line 474: "organization of domains was more similar across conditions within an animal than for the same condition across animals". See Ejaz et al. (2015). Nature Neuroscience, for a similar observation in human fMRI motor maps.

We agree and now cite Ejaz et al. (2015) on line 315.

Line 483: The overlap analysis is unfortunately quite dependent on the threshold.

overhauled our thresholding procedure so that it is now based on a t-test that includes all trials. We believe that the thresholded maps are now considerably more robust than in the original submission. We provide more detail in our response to the first major point from the same reviewer.

Line 494: form → from?

Revised.

I am not sure what I am supposed to learn from the cluster analysis. It is entitled "Clustering time courses recapitulates spatial patterns of activity" – likely referring to the similarity between Figure 6a and 6b. As the clustering partly depends on the magnitude of the signal at the time points presented in Figure 6a, what is the alternative? Is this similarity more than expected by chance?

The clustering analysis has been removed from the manuscript.

Nevertheless, yes, similarities between Figure 6A and 6B drove our conclusion that clustering returned maps that were comparable to the activity maps from averaging frames. It is important to note that clustering was conducted on the 70 time points that make up a time series (Figure 6B and 6E in original submission), whereas activity maps were averages from just five time points (Figure 6A and 6D in original submission).

Indeed, we believe that the similarities were more than expected by chance. We estimated the noise ceiling from running the clustering analysis after shuffling the timecourse of every pixel. This perturbation changed the temporal profile of every pixel but preserved signal magnitude and spatial relationships between pixels. Clustering the shuffled time courses returned salt and pepper maps that had no resemblance to original Figure 6A-E. We should have included this control in the original submission.

Line 527-546. The overlap between the maps is very hard to interpret, as it is highly dependent on the threshold procedure applied, which is somewhat arbitrarily chosen (see above). Even random maps of a certain size would overlap to some degree, and even identical "true" maps don't overlap completely due to measurement error. An overlap of 0.5 has no intrinsic meaning. It would probably be more informative to report the correlation between unthresholded maps, and compare this to a noise ceiling, where you correlate the maps of one condition across sessions.

We agree and have overhauled the analysis in question. Conditions are now compared in two ways. First, direct statistical comparison between the reach-to-grasp condition and the reach-only condition (Figure 6B-C). Second, correlating average time series from pairs of conditions (Figure 6D). Both comparisons reported differences between the precision grip condition and the reach-only condition. Those two conditions had the largest difference in forelimb joint kinematics and muscle activity (Figure 7).

Reviewer #2 (Recommendations for the authors):

Chehade and Gharbawie investigated motor and premotor cortex in macaque monkeys performing grasping and reaching tasks. They used intrinsic signal optical imaging (ISOI) covering an exceedingly large field-of-view extending from the IPS to the PS. They compared reaching and fine/power-grip grasping ISOI maps with "motor" maps which they obtained using extensive intracranial microstimulation. The grasping/reaching-induced activity activated relatively isolated portions of M1 and PMd, and did not cover the entire ICM-induced 'motor' maps of the upper limbs. The authors suggest that small subzones exist in M1 and PMd that are preferentially activated by different types of forelimb actions. In general, the authors address an important topic. The results are not only highly relevant for increasing our basic understanding of the functional architecture of the motor-premotor cortex and how it represents different types of forelimb actions, but also for the development of brain-machine interfaces. These are challenging experiments to perform and add to the existing yet complementary electrophysiology, fMRI, and optical imaging experiments that have been performed on this topic – due to the high sensitivity and large coverage of the particular IOSI methods employed by the authors. The manuscript is generally well written and the analyses seem overall adequate – but see below for some additional analyses that should be done. Although I'm generally enthusiastic about this manuscript, there are two major issues that should be clarified. These major questions relate mainly to potential thresholding issues and clustering issues.

1) The main claim of the authors is that specific forelimb actions activate only a small fraction of what they call the motor map (i.e., those parts of M1/PMd that evoke muscle contractions upon ICM). The action-related activity is measured by ISOI. When looking a the 'raw' reflectance maps, it is rather clear that relatively wide portions of the exposed cortex are activated by grasping/reaching, especially at later time points after the action. In fact, another reading of the results may be that there are two zones of 'deactivation' that split a large swath of motor-premotor cortex being activated by the grasping/reaching actions. (e.g. at 6 seconds after the cue in Figure 3A, 5A). At first sight, the 'deactivated' regions seem to be located in the cortex representing the trunk/shoulder/face – hence regions not necessarily activated (or only weakly) during the grasping/reaching actions. If true, this means that most of the relevant M1/PMd cortex IS activated during the latter actions – opposing the 'clustering' claims of the authors. This raises the question of whether the 'granularity' claimed by the authors is: a. Threshold dependent. In this context, the authors should provide an analysis whereby 'granularity' is shown independent of statistical thresholds of the ISOI maps.

We appreciate the reviewer’s concerns and have completely revised the analyses central to Figure 5. We believe that the figure now contains evidence from both thresholded and unthresholded ISOI data in support of limited spatial extent of cortical activation (i.e., “granularity” in the reviewer’s comments).

For evidence from unthresholded ISOI data, we examined reflectance change time courses from different size ROIs (line 764-768). (A) Small circular ROIs (0.4 mm radius), which we placed in the M1 hand, M1 arm, and PMd arm, zones (Figure 5B). (B) Large ROI inclusive of the M1 and PMd forelimb representations (Figure 5B). We reasoned that if cortical activity is spatially widespread, then the small and large ROIs would report similar time courses. In contrast, if cortical activity is spatially focal, then activity would be detected in the small ROI time courses but would washed out in the large ROI time courses. Our results support the second possibility (Figure 5C-F). Thus, in the movement conditions, time courses from the small ROIs had a large negative peak after movement completion (Figure C-E). In contrast, the characteristic negative peak was absent in the time courses obtained from the large ROI (Figure 5F).

Separately, we revised our thresholding approach to make those results less sensitive to thresholding effects (more details in our response to the first major point from Reviewer 1). The revised results – thresholded/ binarized maps – are consistent with focal cortical activity. Figure 5G & 5J show activity maps thresholded (t-test, p<0.0001) without correction for multiple comparisons, and therefore represent the least restrictive estimate of the spatial extent of cortical activity. Measurements from these maps showed that significantly active pixels overlapped <40% of the M1 & PMd forelimb representations. We interpret the thresholded results as evidence in support of focal cortical activity.

b. Dependent on the time-point one assesses the maps. Given the sluggish hemodynamic responses, it is unclear which part of the ISOI maps conveys the most information relative to the cue and arm/hand movements. I suspect that timepoints > 6 s will reveal even larger 'homogeneous' activations compared to the maps < 6s.

We agree with the reviewer that the lag in hemodynamic signals complicates frame selection. Nevertheless, it is unlikely that cortical activity maps would have been larger at time points >6s from Cue. We provide three supporting arguments.

(1) In the imaging sessions used in Figure 4, we acquired images for 9s per trial and systematically varied Cue onset time. The time courses in Figure 4A-B show that for all Cue onset conditions, the negative peak occurred <6s from Cue. This observation from unthresholded results does not support the notion of greater cortical activity at time points >6s from Cue.

(2) From the same experiment, Figure 4C shows 9 thresholded/binarized maps generated from different time points in relation to Cue. We measured the size of each map (i.e., overlap with the M1/PMd forelimb representations). We present the results in a scatter plot in Author response image 2. The largest maps came from an average frame captured +5.8-6.0s from Cue. Those maps are on the diagonal in Figure 4E (top left to bottom right). This result from thresholded data therefore does not support the notion of greater cortical activity at time points >6s from Cue.

Author response image 2

Download asset Open asset

(3) In all other sessions, we acquired images for 7s per trial (-1.0 to +6.0 s from Cue) without varying Cue onset time. At every time point (100 ms), we measured the size of the thresholded/binarized map in relation to the size of the M1 and PMd forelimb representations. The results are presented in Figure 5I & 5L and indicate that thresholded maps plateau in size by 5.0-5.5 s from Cue. At peak size, the maps overlapped <50% of the M1 and PMd forelimb representations. These result indicates that it is unlikely that we underreported the size of activity maps by not measuring map size beyond 6s from Cue.

In fact, Figure 5F (which is highly thresholded) shows a surprisingly good match between the different forelimb actions, which argues against the existence of small subzones that are preferentially activated by different types of forelimb actions -the main claim of the authors.

Our original proposal should have been more clearly stated. We were proposing that the thresholded maps, which had similar spatial organizations across conditions as the reviewer suggested, reported on subzones tuned for reach-to-grasp actions. Adjacent to those subzones could be other subzones that are preferentially active during other types of forelimb actions (e.g., pulling, pushing, grooming). We could not test this possibility in our study because the behavioral task examined a narrow range of arm and hand actions. We therefore revised the Discussion to state the limitations of our task and to lean more on published work that supports the present proposal (439-443 and 504-508).

2) Related to the previous point, the ROI selections/definitions for the time course analyses seem highly arbitrary. As indicated in the introduction, the clustering hypothesis dictates that "an arm function would be concentrated in subzones of the motor arm zones. Neural activity in adjacent subzones would be tuned for other arm functions." To test this hypothesis directly in a straightforward manner, the authors could use the results from the ICM experiment to construct independent ROIs and to evaluate the ISOI responses for the different actions. In that case, the authors could do a straightforward ANOVA (if the data permits parametric analyses) with ROI, action, and time point (and possibly subject) as factors.

We agree with the reviewer, and we now leverage the ICMS map for guiding ROI placement. All time courses are now derived from 1 of 2 types of ROIs. (1) Small ROIs (0.4 mm radius) placed in zones defined from ICMS (e.g., M1 hand zone). (2) Large ROIs that include the entire forelimb representations in M1 or in PMd (Figure 5B).

3) More details about the transparent silicone membrane should be provided: How implanted? How maintained? Please provide pictures at the end of the 9 and 22months periods.

We expanded the Methods section to provide additional information on the silicone membrane (line 566-577).

The optical imaging chamber and the silicone membrane were not implanted until a monkey was considered trained on the task (9 and 22 months, respectively). We revised the Methods to clarify this point (line 566-567).

4) Figure 1D: Are the yellow circles task cues that the monkeys saw? If not, what were the different cues (Reach, Grasp, Withhold)? Were precision and power grip (and reach?) trials similarly cued?

The yellow circles were digitally added to the still frames to make the target obvious to the reader. The monkey did not see the yellow circles. We revised the figure legend to clarify this point (line 1005).

Identical Go Cues were provided for the 3 movement conditions (i.e., reach-to-grasp large sphere, reach-to-grasp small sphere, and reach only). The purpose of the Go Cue was to instruct the monkey to start reaching. In any given trial, the target object was visible to the monkey throughout the trial. Thus, the type of movement condition was not cued as it was self-evident at the outset of the trial. We revised the Methods to ensure that there is no confusion about Cues (line 583-585; 593-594; 611-612).

5) For the MEG activity a total of 14 27gauge wires were inserted in 7 muscles. This sounds rather invasive. Did monkeys tolerate this easily? Were local anesthetics used? How deep were the wires inserted in the muscle? How did one prevent the monkeys grabbing these wires?

EMG recordings were conducted in separate sessions from ISOI. Once the monkey was head-fixed, we administered a single, low dose of ketamine (2-3 mg/kg, IM). The point of this mild sedation was to minimize any discomfort for the animal during EMG wire insertion, and to create a safe experience for the animal and the handler. The sedation achieved both objectives without eliminating reflexes or muscle tone. We have no reason to believe that the monkey did not readily tolerate the wire insertion procedure or the EMG recordings. Behavioral testing and EMG recordings started 45-60 minutes after sedation. By that point, there were no outwardly lingering effects of sedation and the monkey rapidly reached and grasped treats from the experimenter (line 633-643). Also, there were no differences in reaction time, reach speed, or success rate, on sessions without mild sedation and sessions with mild sedation and EMG recordings.

EMG wires were inserted ~15 mm below the surface of the skin.

Only the non-working forelimb could have disturbed the EMG wires that we inserted into the working forelimb. But the non-working forelimb was restrained from the outset. This point is now in the Methods (line 642-643).

6) The large FOV was illuminated using 2-3 LEDs. Was illumination uniform throughout the FOV? If not, can this lead to inhomogeneous sensitivity?

Even illumination is essential for high quality ISOI. We invested effort at the start of every session to optimize illumination. We revised the Methods (line 730-732) to include two factors that were critical to achieving even illumination. (1) Diffuser lens attached to each LED. (2) Real-time heat map of the field-of-view for positioning the LEDs and adjusting the intensity of each one.

7) Pixel values were clipped, if I understood correctly, at either 0.3 SD or 1 SD from the mean. This significantly changes the dynamic range of these pixel values. Are results different without clipping?

Pixel values were clipped only to facilitate visualization of optical images (e.g., Figure 3A, 5A, 6A). No clipping was done wherever quantitative measurements were involved (e.g., Figure 5G-L). Thus, there was no clipping for time courses, t-tests, and spatial measurements. We revised the Methods to clarify this point (line 759-761).

8) An arbitrarily p<0.0001 statistical threshold was used. Why was no correction for multiple correspondence performed as one collects data from ~ 600k-1400k pixels (yet without considering the masks)?

We redid the analyses after Bonferroni correction for multiple comparisons (p<0.1e-7; 0.05/350k pixels [excluding masked pixels]). We describe the correction in Methods (line 791-794). We now report results thresholded at p<1e-4 and thresholded at p<1e-7 (Figure 5G-L; Figure 5 – Supplemental Figure 1). Not surprisingly, maps were smaller in the more conservative threshold.

We note that many labs consider the Bonferroni correction an overly conservative approach for correcting family wise errors in imaging. The correction assumes that pixels (or voxels in fMRI) provide independent measurements, which is not the case given the close functional relationship between every pixel with its neighbors. An alternative approach would have been to determine the number of multiple comparisons from the number of ROIs. But we had only 4 ROIs (Figure 5B), which would have resulted in a threshold (p<0.01, [p=0.05/4]) that is more lenient than the one adopted here and in the original submission (p<0.0001).

9) What is the reason for obvious edge effects in the withhold condition (Figure 3A, 3F) and apparently also the precision grip condition (Figure 5A: 1.1-3.5sec)? Is this artifactual?

We understand the reviewer’s concerns, which were also raised by the first reviewer. We believe that the seeming edge effect is activity in major blood vessels. Specifically, the posterior border of the field-of-view coincides with the central sulcus, which contains a major blood vessel. The darkening is most likely due to reflectance change in the vessel as opposed to an edge effect of the cranial window. Similarly, the rostral edge effect coincides with the arcuate sulcus. We now include this explanation in the Results (line 157-160).

10) How are the ROIs defined (cfr major point 2)? They differ in size and location between analyses (e.g. Figure 3 vs Figure 5/S1). Please discuss.

We agree with the reviewer that the ROIs differed in size and location across analyses. We now use 1 of 2 types of ROIs. (1) Circular ROI that is the same size across analyses. (2) ROI fitted to the forelimb representation in M1 or PMd. The shape and size of those ROIs is guided exclusively by the motor map. We revised the Methods to reflect these points (line 764-768).

11) I find it a bit counter-intuitive that the same color-code (i.e. black values) is given for 'activations' in panel 3G and 'deactivations' in panel 3H. Why not using red and blue instead? Along the same vein: activations in the time-courses are negative (reflecting the darkening). It would be instructive for the non-experienced reader to either add activation/ suppression on the figures (above and below zero in Figures 3I and J and 5C), or to invert the Y-axis.

We agree with the reviewer on all accounts. We revised the panels in Figure 3 so that significant pixel darkening is shown in red and significant pixel brightening is shown in blue. Separately, we revised the y-axis labels for all time course plots to facilitate interpretation of negative and positive values.

12) Line 426: Figure 5K does not exist.

We corrected the error.

13) Line 520: It is argued that the unsupervised cluster analysis, which is quite interesting, is similar in both monkeys. However, this is not obvious from the data: neither in the spatial domain (Figure 6B vs S2B) nor in the temporal domain Figure 6C and S2 C, especially the blue plots. In fact, the clustering data from monkey G reveal a rather widespread, uninterrupted pattern arguing against the 'cluster hypothesis' of the authors. This should be discussed in more depth.

We agree with the reviewer’s concerns. As stated in response to reviewer 1, we replaced the clustering analysis with a more straightforward analysis that we now present in Figure 6 (line 345-374).

14) Figure 7A: It is unclear which analysis was done. Is this simply giving 3 different colors to each pixel – indicating 1) precision > baseline; 2) reach > baseline 3) both > baseline (for the left panel)? Why not performing straight (pair-wise) subtractions?

We agree that the analysis and the presentation of the results were both confusing. We replaced the analysis with pairwise comparisons using t-tests and cross-correlations (Figure 6, line 345-374).

Author details

1. Nicholas G Chehade

   1. Department of Neurobiology, University of Pittsburgh, Pittsburgh, United States
   2. Systems Neuroscience Center, University of Pittsburgh, Pittsburgh, United States
   3. Center for the Neural Basis of Cognition, Pittsburgh, United States
   4. Center for Neuroscience, University of Pittsburgh, Pittsburgh, United States

   Contribution

   Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-1296-948X

2. Omar A Gharbawie

   1. Department of Neurobiology, University of Pittsburgh, Pittsburgh, United States
   2. Systems Neuroscience Center, University of Pittsburgh, Pittsburgh, United States
   3. Center for the Neural Basis of Cognition, Pittsburgh, United States
   4. Center for Neuroscience, University of Pittsburgh, Pittsburgh, United States
   5. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States

   Contribution

   Conceptualization, Resources, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing – original draft, Project administration, Writing – review and editing

   For correspondence

   omar@pitt.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-2744-9305

• 622

  Page views

• 125

  Downloads

• 0

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.