Preparation is an important component of voluntary movement. Throughout the motor system, many neurons show modulation of firing rate that occurs during movement preparation and is linked to parameters of the ensuing movement (Dorris et al., 1997; Glimcher and Sparks, 1992; Hanes and Schall, 1996; Munoz and Wurtz, 1995; Tanaka and Fukushima, 1998; Tanji and Evarts, 1976; Weinrich and Wise, 1982; Wurtz and Goldberg, 1972; Churchland et al., 2006a; Churchland et al., 2006b; Darlington et al., 2018; Hanes and Schall, 1996; Mahaffy and Krauzlis, 2011; Messier and Kalaska, 2000). Importantly, many neurons that seem to be tied to movement still have preparatory activity, even though there is neither movement nor muscle activity during preparation. How can motor and premotor cells in the cerebral cortex and even the brainstem and spinal cord (Glimcher and Sparks, 1992; Munoz and Wurtz, 1995; Prut and Fetz, 1999) show preparatory activity without causing inappropriate movement?

We have addressed the general question of how preparatory activity fails to evoke movement in the smooth eye movement region of the frontal eye fields (FEF_SEM), a critical node in the neural circuit for smooth pursuit eye movements. Neurons in FEF_SEM are seemingly tied to pursuit. Inactivation or lesioning of FEF_SEM produces dramatic defects in smooth pursuit (Keating, 1991; Shi et al., 1998). Micro-stimulation induces smooth eye movement (Gottlieb et al., 1993). Strong modulation of activity occurs just before the initiation of pursuit (MacAvoy et al., 1991). The modulation is tuned for the direction of visual motion and eye movement and is monotonically related to their speeds (Gottlieb et al., 1994; Tanaka and Lisberger, 2002a; Tanaka and Lisberger, 2002b). Therefore, the activity of FEF_SEM cells seems to play an important role in the generation of smooth pursuit eye movement. Importantly, the same cells that discharge in relation to the initiation of pursuit also show an impressive ramp increase in firing rate during fixation, at a time when there is neither visual motion nor smooth eye movement (Mahaffy and Krauzlis, 2011; Tanaka and Fukushima, 1998). The preparatory activity grows over the hundreds of milliseconds preceding visual motion and the onset of pursuit, and its amplitude encodes expectation of upcoming target speed and combines with visual-motion input in a reliability-weighted manner to determine pursuit-related responses (Darlington et al., 2018).

Several mechanisms have been proposed in various motor systems to explain how preparatory activity can evolve without causing movement. In the arm movement system, the current paradigm suggests that populations of neurons act as a dynamical system in which preparatory activity sets the initial conditions for the progression of movement activity (Churchland et al., 2010; Churchland et al., 2012). Importantly, preparatory population activity in primary motor cortex (M1) and dorsal premotor cortex (PMd) progresses in a way that avoids dimensions related to movement (Kaufman et al., 2014). It remains confined to ‘movement-null’ dimensions until just before movement initiation, and only then begins to progress along ‘movement-potent’ dimensions. Indeed, the preparatory and movement-related population responses in M1 and PMd occupy orthogonal dimensions (Elsayed et al., 2016). Furthermore, the use of output-null neural dimensions plays a role in both M1’s initial processing of visuomotor feedback during movement as well as short-term force-field adaptation in PMd (Perich et al., 2018; Stavisky et al., 2017).

In the saccadic eye movement system, movement is prevented during preparation via the action of an all-or-none inhibitory gate created by the action of omnipause neurons in the brainstem (Cohen and Henn, 1972; Keller, 1974; Luschei and Fuchs, 1972). Ramps of preparatory activity in the saccadic region of the frontal eye fields fail to trigger eye movements until the collective descending activity from the frontal eye fields and superior colliculus cause the omnipause neurons to cease firing, releasing the circuits that generate movement (Dorris et al., 1997; Hanes and Schall, 1996). Furthermore, subthreshold activity in the superior colliculus possesses some motor potential when omnipause neurons are inhibited by the trigeminal blink reflex (Jagadisan and Gandhi, 2017). However, there is no evidence that the pursuit system uses the all-or-none gate that controls saccades (Missal and Keller, 2002; Schwartz and Lisberger, 1994).

Here we propose a third, different, mechanism for movement prevention during motor preparation. The outputs from the FEF_SEM seem to control the gain of visual-motor transmission for pursuit (Nuding et al., 2009; Tanaka and Lisberger, 2001). Further, visual-motor gain is dialed up in preparation for an imminent smooth eye movement (Kodaka and Kawano, 2003; Tabata et al., 2006). In the present paper, we show striking parallels between the progression of preparatory activity in FEF_SEM and the preparatory enhancement of visual-motor gain in the pursuit system. We argue that the output of FEF_SEM dials up visual-motor gain in preparation for a behaviorally-relevant visual motion of a specific direction and speed, and that the use of FEF_SEM output as a sensorimotor gain signal allows preparation to proceed without causing movement. We also show that preparatory activity in FEF_SEM seems to contradict the major predictions of the dynamical system framework that has arisen from work in arm motor cortex (Churchland et al., 2010; Elsayed et al., 2016; Kaufman et al., 2014).

We present three primary findings to support the conclusion that FEF_SEM preparatory activity can evolve without causing smooth pursuit eye movement because its output does not drive smooth eye movement, but rather is used as a gain signal that controls access of visual motion to the oculomotor machinery. First, we demonstrate parallel progression of FEF_SEM preparatory activity and an objective measure of visual-motor gain. Second, we show that preparatory activity in FEF_SEM evolves in a manner that predicts movement when there is none, precluding for the smooth pursuit eye movement system the ‘movement-null’ space mechanism proposed for M1 and PMd preparatory activity in the arm movement system (Kaufman et al., 2014). Third, we show that there is some degree of reorganization of the FEF_SEM population response, and we link the reorganization to a directionally nonspecific component of preparatory modulation of visual-motor gain. We suggest that gain control should be considered in all movement systems as one of several possible mechanisms to allow preparatory activity without movement.

FEF_SEM preparatory activity evolves in parallel with a preparatory modulation of visual-motor gain

Neurons in FEF_SEM are seemingly tied to the generation of smooth pursuit eye movement, yet display preparatory activity in the absence of any eye movement. We recorded single neurons in FEF_SEM while monkeys smoothly tracked moving targets during blocks of trials that presented target motion in a single direction. The example recording in Figure 1A–C shows the response of a neuron with a strong ramp increase in firing rate during fixation of a stationary spot, well in advance of either (1) target motion that starts 800–1600 ms after fixation onset or (2) the initiation of eye movement. The same neuron shows a second increase in firing rate that starts after the onset of target motion, around the time of the initiation of smooth pursuit (arrow in Figure 1C). We observed a wide range of preparatory- and pursuit-related firing rate modulation across our FEF_SEM population, and there was a positive relationship between preparatory- and pursuit-related firing rate modulation (Figure 1D, Monkey Re: Spearman’s correlation coefficient = 0.17, p=0.02, n = 187 data points from 80 neurons; Monkey Xt: Spearman’s correlation coefficient = 0.55, p=4.5×10⁻¹⁰, n = 110 data points from 66 neurons). As explained in the Materials and methods, we ran multiple blocks of trials on most neurons using different directions of target motion, and each block provided a data point. Thus, more data points than neurons.

Figure 1

Download asset Open asset

Figure 1—source data 1

Source data for Figure 1.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig1-data1-v2.mat

Download elife-50962-fig1-data1-v2.mat

We hypothesized that preparatory activity in FEF_SEM fails to cause movement because it is used downstream as a visual-motor gain signal. Gain can be dialed up by preparatory activity, but absent a movement-driving visual motion signal there will be no movement. This hypothesis is based on previous publications showing that FEF_SEM output during pursuit initiation is involved in controlling the gain of visual-motor transmission (Tanaka and Lisberger, 2001).

To test whether FEF_SEM preparatory activity could be acting as a gain signal, we conducted experiments that measured only pursuit behavior, but were carefully contrived based on knowledge of the properties of preparatory activity in FEF_SEM. We probed the state of visual-motor gain in pursuit by delivering a brief (50 ms) pulse of target motion at 5 deg/s during fixations leading up to single-direction pursuit trials. We delivered the pulse at different times during the build-up of preparatory activity and in different states of preparation. Because the test pulses of visual motion were always exactly the same and eye speed was zero at the time of the pulse, any effect of fixation time or preparatory state on the eye movement responses to the probe must reflect a change in the setting of visual-motor gain in the pursuit system.

It is important to note that the visual motion pulse experiments measured eye movement behavior only and were a direct follow-up from the analyses we performed on our neurophysiological data in FEF_SEM. We matched all experimental conditions, including the timing of fixation and visual motion onset, in order to maximize the validity of comparisons between the behavioral and neurophysiology data sets. Further, the neurophysiological and behavioral experiments were conducted on the same pair of monkeys.

We found a series of parallels between the amplitude of preparatory activity and the setting of visual-motor gain, as probed by the size of the eye movement response to the pulse. Parallels include: (1) time course during fixation before target motion, (2) effect of variations in the expectation of target speed, and (3) direction specificity.

The behavioral assessment of visual-motor gain increases as a function of fixation time, paralleling the progression of preparatory activity in FEF_SEM. Figure 2A illustrates data from one example session showing larger eye speed responses at later fixation times in response to identical, but differently-timed, pulses of visual motion. The same result appeared consistently in both monkeys we tested across multiple behavioral sessions (Figure 2B). Modulation of FEF_SEM preparatory activity showed a very similar time course across fixation for the overwhelming majority of neurons that had positive preparatory firing rate modulation (Figure 2C, Figure 2—figure supplement 1). We can see the tight link between visual-motor gain and preparatory firing if we plot the behavioral results from Figure 2B against the preparatory firing from Figure 2C for each fixation time and each monkey (Figure 2D). The relationship is fairly linear with a positive correlation between neural responses and behavior across our population of neurons (slopes were computed for each neuron using linear regression: monkey Re mean = 0.073 degrees/spike, p=2.34×10⁻¹¹, n = 135 data points from 66 neurons monkey Xt mean = 0.074 degrees/spike, p=2.76×10⁻¹¹, n = 67 data points from 45 neurons).

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Source data for Figure 2.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig2-data1-v2.mat

Download elife-50962-fig2-data1-v2.mat

The behavioral assessment of visual-motor gain varies in parallel with preparatory activity when we change the expected target speed by changing the blend of target speeds in a block of trials. We showed previously that the amplitude of FEF_SEM preparatory activity encodes an expectation of target speed based on experience in blocks of 50 trials with different blends of target speed (Figure 3A; Darlington et al., 2018). Preparatory firing rate modulation was larger in magnitude (Figure 3B) during the ‘fast-context’ (green traces, 80% of trials at 20 deg/s and 20% at 10 deg/s) compared to during the ‘slow-context’ (blue traces, with 80% of trials at 2 deg/s and 20% at 10 deg/s). Based on behavioral assessment using pulses of visual motion during fixation, visual-motor gain during the preparatory period also is enhanced in the fast context compared to the slow context. Eye velocity responses were larger during the fast-context versus the slow-context when we presented the same visual motion pulses during fixation at different times (green versus blue in Figure 3C, scatter plot in Figure 3D). In the fast-context versus slow-context, the mean eye velocity responses to the pulse were 3.02 ± 0.71 deg/s (SD) versus 2.27 ± 0.73 deg/s (SD), and the difference was statistically significant (p=8.30×10⁻⁶, paired, two-sided Wilcoxon signed rank test, n = 26 behavioral experiments in two monkeys, z = 4.46, Cohen’s d = 1.06).

Figure 3

Download asset Open asset

Figure 3—source data 1

Source data for Figure 3.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig3-data1-v2.mat

Download elife-50962-fig3-data1-v2.mat

The directional specificities of preparatory modulation of visual-motor gain and FEF_SEM preparatory activity vary in parallel. To assess the directional specificity of preparatory visual-motor gain modulation, we delivered visual motion pulses during fixation again, but now in four different directions relative to the consistent direction of the pursuit target motion in a given single-direction block: 0 deg, ‘same-direction’;±90 deg, ‘orthogonal-direction’; 180 deg, ‘opposite-direction’. Across the full duration of fixation, eye speed responses were largest for same-direction pulses, intermediate for orthogonal-direction pulses, and smallest for opposite-direction pulses (Figure 4A). For all directions, eye speed responses increased as a function of time during fixation for same-, orthogonal-, and opposite-direction pulses. We quantified the effect of fixation time by computing the slope of response amplitudes across pulse times. For the same direction, orthogonal direction, and opposite direction pulses, the mean (+/- S.D.) slopes across 30 experiments in two monkeys were 1.05 ± 0.59 deg/s per s, 0.73 ± 0.40 deg/s per s, and 0.72 ± 0.45 deg/s per s. The effect of fixation time on response amplitude was significant across all directions using 2-sided Wilcoxon signed rank tests: same-direction (p=2.60×10⁶, z = 4.70); orthogonal-direction (p=1.73×10⁻⁶, z = 4.78); and opposite-direction (p=4.78×10⁻⁶, z = 4.78). The differences from same-direction were statistically significant using the paired sample, 2-tailed Wilcoxon signed rank test for n = 30 experiments on two monkeys: same versus orthogonal-pulses (p=0.0093, z = 2.60, Cohen’s d = 0.63); same- versus opposite-direction pulses (p=0.029, n = 30, z = 2.19, Cohen’s d = 0.62). Thus, there is a greater preparatory enhancement of visual-motor gain in the expected direction of the upcoming pursuit trial. Further, the increase in response amplitude over fixation time for all directions of pulses indicates a degree of directionally non-specific preparatory enhancement of visual motor gain.

Figure 4

Download asset Open asset

Figure 4—source data 1

Source data for Figure 4.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig4-data1-v2.mat

Download elife-50962-fig4-data1-v2.mat

FEF_SEM preparatory activity also is related to the expectation of target direction created by adjusting the preponderance of target directions in a block of trials. We divided our FEF_SEM population into three groups according to the relative difference between the preferred direction of each cell and the direction of the experiment: same direction, 0 ± 45 deg; orthogonal, +90 ± 45 deg and −90 ± 45 deg; and opposite, 180 ± 45 deg. Consistent with a larger behavioral preparatory enhancement of visual-motor gain in the expected direction, FEF_SEM neurons display the largest modulation of firing rate during fixations leading to target motion towards their preferred direction (Figure 4B). Preparatory firing rate modulation was smallest for upcoming target motions that were opposite to the preferred direction and intermediate for upcoming target motions that were orthogonal to the preferred direction.

We find even more compelling neural correlates of the behavior shown in Figure 4A using a semi-supervised variant of principal component analysis: demixed principal component analysis (dPCA) (Kobak et al., 2016). The primary difference is that dPCA retains knowledge of experimental conditions when finding components, allowing for identification of different patterns of population activity that are independent of, or dependent on, experimental condition. We performed dPCA on the preparatory responses of a subset of our FEF_SEM population (n = 63 neurons) from which we were able to collect data during experiments toward and away from the preferred directions of each neuron. Thus, the sole condition in our dPCA is the direction of the ensuing target motion and eye movement. Our aim was to identify ‘direction-dependent’ and ‘direction-independent’ components to test the idea that FEF_SEM contributes to the directionally specific and non-specific components of preparatory visual-motor gain enhancement shown in Figure 4A.

FEF_SEM population activity can fully account for the directional features of preparatory visual-motor gain enhancement. The first three components obtained via dPCA account for ~90% of the variance of our FEF_SEM population preparatory response (Figure 4C). Each of the three primary components predict features of the behavioral data shown in Figure 4A. Component one is direction-independent. The projections of FEF_SEM preparatory population activity onto component one both increase across fixation for impending target motion toward and away from the preferred direction (Figure 4D). The projections onto component one are consistent with the directionally nonspecific enhancement of visual-motor gain observed for the orthogonal and opposite-direction visual motion pulses. Component two is direction-dependent. There are two primary differences between the projections of FEF_SEM population activity onto component two during preparation for target motion toward and away from the preferred direction (Figure 4E). (i) The projections are offset from each other with the projections for toward and away motion sitting above and below each other, respectively. The offset between the projections is consistent with the behavioral finding of an offset between the eye speed responses to brief target motions in the same versus opposite direction from expected target motion: there is a substantial difference in eye speed responses to the same and opposite visual motion pulses even at the earliest fixation times we tested. (ii) The projection for motion toward the preferred direction increases and that for motion away from preferred projection decreases during the latter half of the fixation period. Component three also is direction dependent. The projection of FEF_SEM population activity onto component three also shows a ramp increase for impending target motion toward the preferred direction and a ramp decrease for impending target motion away from the preferred direction. The increase and decrease of the projections onto components 2 and 3 are consistent with the larger enhancement of visual-motor gain across fixation time for the same versus opposite direction visual motion pulses. In conclusion, the neural basis for the behavioral preparatory enhancement of visual-motor gain can be fully reconstructed from the first 3 dPCA components of the FEF_SEM preparatory population response.

Modulation of FEF_SEM preparatory activity occurs along movement-potent dimensions

Studies of arm movement have shown that population preparatory activity in M1 and PMd evolves in a manner that does not predict movement (Kaufman et al., 2014). Movement is prevented during preparation because the population activity resides in ‘movement-null’ neural dimensions. We now test this idea on our population of FEF_SEM neurons. A priori, such a mechanism could generalize to the smooth pursuit eye movement system, especially since humans and monkeys can initiate smooth pursuit in the absence of visual motion under certain conditions (Freyberg and Ilg, 2008; de Hemptinne et al., 2006; Kowler et al., 2019). Even if the output of FEF_SEM truly is used as a gain signal, noise immunity might be enhanced if population activity resided in an output null state until there is a believable visual motion signal to drive pursuit initiation. Two mechanisms for preventing unwanted movement might be better than one.

We established the movement-potent dimensions for our sample of FEF_SEM neurons from blocks of trials where the monkey pursued a patch of dots that moved in each of 8 randomly chosen directions (Figure 5A). The population average of FEF_SEM firing across our full sample of neurons was strongly tuned to the direction of target motion and pursuit (Figure 5B), with a large increase in firing rate in the ‘preferred direction’ of the cell, a smaller increase when motion was at +/- 45 degrees from the preferred direction, no modulation for visual motion orthogonal to the preferred direction, and decreases in firing rate for +/- 135 degrees and 180 degrees from the preferred direction. We represented the movement-potent dimension for our population by computing the optimal linear model that related the pursuit-related firing of our sample of FEF_SEM neurons to the eye movements evoked during the initiation of pursuit in eight directions. The linear model optimized two weights for each neuron: one related the neuron’s firing rate to horizontal eye velocity and the second related the neuron’s firing rate to vertical eye velocity. Our fitting procedure assigned a continuous distribution of weight magnitudes to the members of our population (Figure 5C), showing that the model used the entire FEF_SEM population and did not simply pick out a few particularly informative cells. The model worked well on data from pursuit trials that were not used to compute the weights: target motion at three different speeds (2, 10, and 20 deg/s) and two different contrasts (12% and 100%) from a separate block of trials that presented visual motion in a single, repeated direction. It produced fairly accurate predictions for horizontal eye velocity (Figure 5D, slope = 1.09, r² = 0.85) and accurately predicted zero vertical eye velocities (data not shown). Because a model based on the responses to 8 directions of motion was able to predict well the responses in a different block of trials with a single direction of motion, we conclude that it is valid to use a single model across different blocks of behavior.

Figure 5

Download asset Open asset

Figure 5—source data 1

Source data for Figure 5.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig5-data1-v2.mat

Download elife-50962-fig5-data1-v2.mat

FEF_SEM preparatory activity does not avoid movement-potent neural dimensions. The optimal linear model predicts movement when there is none, during fixation and preparatory activity. When we passed the trial-averaged population preparatory firing rate modulation from a single-direction block of pursuit trials through the model, we obtained a non-zero prediction for change in eye velocity across fixation (Figure 6A and B). Importantly, the output of the linear model predicts eye movement in a direction that is appropriate for the upcoming target motion according to how neurons across experimental sessions were combined into a pseudo-population (see below): a large positive horizontal component and relatively smaller vertical component.

Figure 6 with 3 supplements see all

Download asset Open asset

Figure 6—source data 1

Source data for Figure 6.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig6-data1-v2.mat

Download elife-50962-fig6-data1-v2.mat

We obtained similar results when we asked whether we could verify the failings of the movement-null hypothesis for FEF_SEM in neural responses from single trials. We generated 1000 single-trial pseudo-population responses by randomly sampling preparatory firing rate modulations from each cell for fixation-duration matched trials across our population. We passed each pseudo-population response into the optimal linear model to predict the change in horizontal and vertical eye speed across fixation, and computed eye direction and speed during fixation on each ‘trial’. Consistent with the findings of Figure 6A and B, the single trial analysis predicts non-zero eye velocity with a significant projection in the direction of the subsequent pursuit trial (Figure 6C and D). The mean predicted direction was −1.3 deg with a 95% bootstrap interval of −65.8 to +71.8 deg; the mean predicted eye speed was 2.9 deg/s. Thus, for the motor system we study, the data do not fulfill one important prediction of the null-space hypothesis: the optimal linear model for pursuit initiation predicts significant non-zero eye velocities driven by preparatory activity.

We made a number of modifications in adapting the analyses performed in Kaufman et al. (2014) to our data. These modifications were made due to a key difference in our behavioral paradigm: the upcoming direction of arm movement was explicitly cued in Kaufman et al. (2014); the direction of the upcoming visual motion was not explicitly cued in our task. Therefore, only the single-direction block allows for accurate preparation, as the monkey is able to anticipate the upcoming visual motion. However, the eight-direction block provides a more rigorous mapping of FEF_SEM pursuit-related firing rate to eye velocity. Thus, we chose to find the optimal mapping of FEF_SEM firing rate to eye movement in the 8-direction block and test preparatory activity in the single-direction block.

Our strategy was then to create as large of a population response as we could, without excluding data points or neurons. We also did not wish to bias our data collection for the single-direction experiment towards any part of the direction tuning curve of our recorded neurons, for example by only running the single-direction experiment in the preferred direction of our neurons. To this end, we ran multiple single-direction blocks of trials for each neuron that we recorded using target directions that were different from each other and at various points on each neuron’s tuning curve. We then rotated each neuron’s direction tuning curves while fitting the linear model to the 8-direction block in a manner that made it seem like all of the single-direction experiments were run in the same direction (0 degrees, rightward). We included multiple single-direction blocks from the same neurons as if they were independent under the relatively safe assumption that there is a different neuron with a tuning curve that is naturally aligned in the manner that we rotated our data.

We performed several control analyses to confirm that none of our analytical choices produced our results. First, the findings of Figure 6 are not a product of oversampling the preferred directions of our FEF_SEM cells during the single-direction experiments (Figure 6—figure supplement 1). Second, we obtained the same results when we used only a single block of trials from each neuron so that our conclusions are strengthened statistically but do not depend on the use of multiple data points from each neuron as independent units (Figure 6—figure supplement 2). Third, a supplementary analysis that is more congruent with the analysis performed in Kaufman et al. (2014), and that uses a model based on the single-direction block, arrives at the same conclusions (Figure 6—figure supplement 3).

Partial reorganization of population activity from preparatory- to pursuit-related dimensions

If preparatory and pursuit-related activity resided in orthogonal dimensions, then we would say that the population activity in FEF_SEM reorganized completely in the transition from preparation to action. If they resided in parallel dimensions, then we would say that they did not reorganize at all. We now show that the truth lies between these two extremes by using principal component analysis to compare the neural dimensions related to preparation and movement. The existence of a partial reorganization explains why we find that preparatory activity predicts movement and still allows the operation of our system to align to some degree with the operation of the arm movement system, where there is a full reorganization.

After reducing the dimensionality of our preparatory- and pursuit-related population responses during the single-direction trial blocks, we analyze the first preparatory and pursuit PCs (PC_Prep1 and PC_Purs1), each of which captures more than 70% of the variance (Figure 7A), the same amount captured by the dimensionality reduction in Elsayed et al. (2016). We validated PC_Purs1 by showing that the projection of pursuit-related activity onto the PC_Purs1 captures differences in both the speed of target motion and the contrast of the target: in parallel with the neural responses, the projections scaled with target speed and are larger in magnitude and shorter in latency for high- compared to low-contrast targets (Figure 7B). As expected, given the results in the previous section using the optimal linear model, the projection of preparatory activity onto PC_Purs1 shows a ramp increase across fixation (Figure 7D). The main difference between the preparatory and pursuit projections is that the preparatory projection ramps up more slowly to a smaller magnitude compared to the pursuit projections.

Figure 7 with 1 supplement see all

Download asset Open asset

Figure 7—source data 1

Source data for Figure 7.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig7-data1-v2.mat

Download elife-50962-fig7-data1-v2.mat

We demonstrate a partial reorganization of population activity between preparation and pursuit-related activity by computing the angle between PC_Prep1 and PC_Purs1. We used a bootstrap approach that sampled trial-averaged neurons from our population, independently found PC_Prep1 and PC_Purs1 for the sampled population, and calculated the angle between them. We used a sampling procedure that created 1000 simulated population responses, each with 160 units having a uniform representation of preferred directions (see Materials and methods and Supplementary Figure 1 for more details). We find that the mean angle between PC_Prep1 and PC_Purs1 is 65.2 degrees and is significantly different from both 0 and 90 (Figure 7C, p<0.001, two-sided bootstrap statistics, 95% bootstrap interval = 61.4–69.2 deg, n = 1000 iterations). Therefore, preparatory- and pursuit-related neural dimensions are neither aligned nor completely orthogonal in our population of FEF_SEM neurons. As with the linear model, our findings were not influenced by our choice to treat multiple data points from each neuron as independent units in our pseudo-population (Figure 7—figure supplement 1).

To understand what features of the neural responses caused the reorganization of the population response, we examined how the PCA component loadings, defined as the weights relating each neuron’s response to the principal component, changed between PC_Prep1 and PC_Purs1. The distribution of the ratios between the PCA component loadings of PC_Prep1 and PC_Purs1, generated from the bootstrap in Figure 3B, is significantly bimodal (Figure 8A, Hartigan’s Dip Test, Dip statistic = 0.0794, p=0.001). One peak is at a positive ratio and a smaller peak is at a negative ratio. A positive ratio indicates that a given unit behaves in a similar manner during preparation and movement, whereas a negative ratio indicates a change in behavior between preparation and pursuit. We conclude that there are two subpopulations in FEF_SEM, and that the one with negative PC_Prep1/PC_Purs1 ratios may be responsible for the partial reorganization of population activity between preparation and movement.

Figure 8

Download asset Open asset

Figure 8—source data 1

Source data for Figure 8.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig8-data1-v2.mat

Download elife-50962-fig8-data1-v2.mat

To test our hypothesis about the role of neurons with negative PC_Prep1/PC_Purs1 ratios, we separated our population into units with positive (Subpopulation 1) and negative (Subpopulation 2) ratios and compared a number of properties of the two subpopulations. (1) Preferred directions (Figure 8B): Subpopulation 1 consistently made up a larger proportion of units in our bootstrap analysis and contains a higher proportion of units with a preferred direction towards the direction of the single-direction experiment. Conversely, Subpopulation 2 contains a higher proportion of units with preferred directions away from the direction of the experiment. (2) Preparatory versus pursuit modulation: When we redid principal component analysis on the full (non-bootstrapped) data, neurons in Subpopulation 1 had statistically larger amplitudes of both preparatory- and pursuit-related modulation (compare red and black data points in Figure 8C; preparatory activity: p=0.019, Cohen’s d = 0.34, pursuit activity: p=2.7×10⁻¹², Cohen’s d = 0.95, 2-sided Wilcoxon rank sum tests). There is a significant positive correlation between preparatory and pursuit modulation in Subpopulation 1 (Figure 8C, Spearman’s correlation coefficient = 0.59, p=7.8×10⁻²¹, n = 207 samples from 130 neurons) and a significant negative correlation in Subpopulation 2 (Figure 8C, Spearman’s correlation coefficient = −0.63, p=3.8×10⁻¹¹, n = 90 samples from 65 neurons). (3) Principal component alignment: For Subpopulation 1 alone, the first principal components of pursuit and preparatory activity are separated by an angle of only 26.2 degrees. For Subpopulation 2 alone, the angle between PC_Prep1 and PC_Purs1 is 151.2 degrees (Figure 8G). These facts suggest that Subpopulation 2 is responsible for the partial reorganization of population activity between preparation and movement, and that the large difference in response amplitudes between the two populations is at least partly responsible for a lack of full reorganization between preparatory- and movement-related dimensions.

The existence of two subpopulations of FEF_SEM neurons allows for a directionally-nonspecific component of preparatory visual-motor gain enhancement. In Subpopulation 1, preparatory modulation of firing rate has conventional direction tuning. It is large and positive when the upcoming target motion will be in the preferred direction of the cell, smaller when it will be orthogonal to the preferred direction, and negative when it will be opposite to the preferred direction of the cell (Figure 8D,E). In Subpopulation 2, preparatory modulation of firing rate has weaker and non-conventional direction tuning. Modulation is positive when the upcoming target motion is orthogonal or opposite to the preferred direction and absent when target motion will be toward the preferred direction (Figure 8D,F). Together, the two subpopulations can explain two features of Figure 4A. The preparatory activity in Subpopulation 1 explains why preparatory enhancement of visual-motor gain is largest in the direction of the upcoming target motion. The smaller preparatory activity in Subpopulation 2 might explain the directionally-nonspecific, and smaller, component of preparatory visual-motor gain enhancement in orthogonal and opposite directions relative to the upcoming target motion.

To further examine differences between the two subpopulations, we repeated the optimal linear model analysis from Figures 5 and 6 separately on each subpopulation. Linear models generated from the two subpopulations perform roughly the same in predicting eye velocity during the single direction block (Figure 9A,D). Subpopulation 1 predicts eye movement during preparation in the same direction as the upcoming visual motion and eye movement (Figure 9B,C). Subpopulation 2 predicts no significant eye movement during preparation (Figure 9E,F). However, it is worth pointing out that the spread around zero for horizontal and vertical eye velocities is quite large. These features are consistent with Subpopulation 2 dialing up visual-motor gain in a directionally-nonspecific manner and they suggest that the nonspecific component of visual-motor gain is fairly uniform across directions.

Figure 9

Download asset Open asset

Figure 9—source data 1

Source data for Figure 9.

https://cdn.elifesciences.org/articles/50962/elife-50962-fig9-data1-v2.mat

Download elife-50962-fig9-data1-v2.mat

In our system, we not only characterize the population activity during preparation and movement, but also understand why it reorganizes in the way that it does and how that reorganization improves the performance of pursuit eye movements. A full analysis of FEF_SEM neural responses during pursuit eye movements shows that (1) there is a partial reorganization of the population between preparation and movement, (2) the reorganization is not complete enough to place the population activity in a movement-null dimension during preparation, (3) output control of the gain of visual-motor transmission rather than of movement kinematics allows the reorganization to be incomplete without causing movement, and (4) the existence of two subpopulations of neurons explains the partial reorganization of population activity and allows the behavior to be preferentially prepared for target motion in an expected direction with a lower level of readiness for all directions of motion.

Source data have been provided for each figure.

1. 1. Bruce CJ
   2. Goldberg ME

   (1985) Primate frontal eye fields. I. single neurons discharging before saccades

   Journal of Neurophysiology 53:603–635.

   https://doi.org/10.1152/jn.1985.53.3.603

   • PubMed
   • Google Scholar
2. 1. Churchland MM
   2. Afshar A
   3. Shenoy KV

   (2006a) A central source of movement variability

   Neuron 52:1085–1096.

   https://doi.org/10.1016/j.neuron.2006.10.034

   • PubMed
   • Google Scholar
3. 1. Churchland MM
   2. Santhanam G
   3. Shenoy KV

   (2006b) Preparatory activity in premotor and motor cortex reflects the speed of the upcoming reach

   Journal of Neurophysiology 96:3130–3146.

   https://doi.org/10.1152/jn.00307.2006

   • PubMed
   • Google Scholar
4. 1. Churchland MM
   2. Cunningham JP
   3. Kaufman MT
   4. Ryu SI
   5. Shenoy KV

   (2010) Cortical preparatory activity: representation of movement or first cog in a dynamical machine?

   Neuron 68:387–400.

   https://doi.org/10.1016/j.neuron.2010.09.015

   • PubMed
   • Google Scholar
5. 1. Churchland MM
   2. Cunningham JP
   3. Kaufman MT
   4. Foster JD
   5. Nuyujukian P
   6. Ryu SI
   7. Shenoy KV

   (2012) Neural population dynamics during reaching

   Nature 487:51–56.

   https://doi.org/10.1038/nature11129

   • PubMed
   • Google Scholar
6. 1. Cohen B
   2. Henn V

   (1972) Unit activity in the pontine reticular formation associated with eye movements

   Brain Research 46:403–410.

   https://doi.org/10.1016/0006-8993(72)90030-3

   • PubMed
   • Google Scholar
7. 1. Darlington TR
   2. Beck JM
   3. Lisberger SG

   (2018) Neural implementation of bayesian inference in a sensorimotor behavior

   Nature Neuroscience 21:1442–1451.

   https://doi.org/10.1038/s41593-018-0233-y

   • PubMed
   • Google Scholar
8. 1. de Hemptinne C
   2. Lefèvre P
   3. Missal M

   (2006) Influence of cognitive expectation on the initiation of anticipatory and visual pursuit eye movements in the rhesus monkey

   Journal of Neurophysiology 95:3770–3782.

   https://doi.org/10.1152/jn.00007.2006

   • PubMed
   • Google Scholar
9. 1. Dorris MC
   2. Paré M
   3. Munoz DP

   (1997) Neuronal activity in monkey superior colliculus related to the initiation of saccadic eye movements

   The Journal of Neuroscience 17:8566–8579.

   https://doi.org/10.1523/JNEUROSCI.17-21-08566.1997

   • PubMed
   • Google Scholar
10. 1. Dum RP
    2. Strick PL

    (1991) The origin of corticospinal projections from the premotor Areas in the frontal lobe

    The Journal of Neuroscience 11:667–689.

    https://doi.org/10.1523/JNEUROSCI.11-03-00667.1991

    • PubMed
    • Google Scholar
11. 1. Economo MN
    2. Viswanathan S
    3. Tasic B
    4. Bas E
    5. Winnubst J
    6. Menon V
    7. Graybuck LT
    8. Nguyen TN
    9. Smith KA
    10. Yao Z
    11. Wang L
    12. Gerfen CR
    13. Chandrashekar J
    14. Zeng H
    15. Looger LL
    16. Svoboda K

    (2018) Distinct descending motor cortex pathways and their roles in movement

    Nature 563:79–84.

    https://doi.org/10.1038/s41586-018-0642-9

    • PubMed
    • Google Scholar
12. 1. Elsayed GF
    2. Lara AH
    3. Kaufman MT
    4. Churchland MM
    5. Cunningham JP

    (2016) Reorganization between preparatory and movement population responses in motor cortex

    Nature Communications 7:13239.

    https://doi.org/10.1038/ncomms13239

    • PubMed
    • Google Scholar
13. 1. Evinger C
    2. Kaneko CR
    3. Fuchs AF

    (1982) Activity of omnipause neurons in alert cats during saccadic eye movements and visual stimuli

    Journal of Neurophysiology 47:827–844.

    https://doi.org/10.1152/jn.1982.47.5.827

    • PubMed
    • Google Scholar
14. 1. Freyberg S
    2. Ilg UJ

    (2008) Anticipatory smooth-pursuit eye movements in man and monkey

    Experimental Brain Research 186:203–214.

    https://doi.org/10.1007/s00221-007-1225-4

    • PubMed
    • Google Scholar
15. 1. Fuchs AF

    (1967) Saccadic and smooth pursuit eye movements in the monkey

    The Journal of Physiology 191:609–631.

    https://doi.org/10.1113/jphysiol.1967.sp008271

    • PubMed
    • Google Scholar
16. 1. Fuchs AF
    2. Kaneko CR
    3. Scudder CA

    (1985) Brainstem control of saccadic eye movements

    Annual Review of Neuroscience 8:307–337.

    https://doi.org/10.1146/annurev.ne.08.030185.001515

    • PubMed
    • Google Scholar
17. 1. Gao Z
    2. Davis C
    3. Thomas AM
    4. Economo MN
    5. Abrego AM
    6. Svoboda K
    7. De Zeeuw CI
    8. Li N

    (2018) A cortico-cerebellar loop for motor planning

    Nature 563:113–116.

    https://doi.org/10.1038/s41586-018-0633-x

    • PubMed
    • Google Scholar
18. 1. Glimcher PW
    2. Sparks DL

    (1992) Movement selection in advance of action in the superior colliculus

    Nature 355:542–545.

    https://doi.org/10.1038/355542a0

    • PubMed
    • Google Scholar
19. 1. Gottlieb JP
    2. Bruce CJ
    3. MacAvoy MG

    (1993) Smooth eye movements elicited by microstimulation in the primate frontal eye field

    Journal of Neurophysiology 69:786–799.

    https://doi.org/10.1152/jn.1993.69.3.786

    • PubMed
    • Google Scholar
20. 1. Gottlieb JP
    2. MacAvoy MG
    3. Bruce CJ

    (1994) Neural responses related to smooth-pursuit eye movements and their correspondence with electrically elicited smooth eye movements in the primate frontal eye field

    Journal of Neurophysiology 72:1634–1653.

    https://doi.org/10.1152/jn.1994.72.4.1634

    • PubMed
    • Google Scholar
21. 1. Griffin DM
    2. Hoffman DS
    3. Strick PL

    (2015) Corticomotoneuronal cells are "functionally tuned"

    Science 350:667–670.

    https://doi.org/10.1126/science.aaa8035

    • PubMed
    • Google Scholar
22. 1. Hanes DP
    2. Schall JD

    (1996) Neural control of voluntary movement initiation

    Science 274:427–430.

    https://doi.org/10.1126/science.274.5286.427

    • PubMed
    • Google Scholar
23. 1. Hikosaka O
    2. Igusa Y
    3. Nakao S
    4. Shimazu H

    (1978) Direct inhibitory synaptic linkage of pontomedullary reticular burst neurons with abducens motoneurons in the cat

    Experimental Brain Research 33:337–352.

    https://doi.org/10.1007/BF00235558

    • PubMed
    • Google Scholar
24. 1. Jagadisan UK
    2. Gandhi NJ

    (2017) Removal of inhibition uncovers latent movement potential during preparation

    eLife 6:e29648.

    https://doi.org/10.7554/eLife.29648

    • PubMed
    • Google Scholar
25. 1. Kaufman MT
    2. Churchland MM
    3. Ryu SI
    4. Shenoy KV

    (2014) Cortical activity in the null space: permitting preparation without movement

    Nature Neuroscience 17:440–448.

    https://doi.org/10.1038/nn.3643

    • PubMed
    • Google Scholar
26. 1. Keating EG

    (1991) Frontal eye field lesions impair predictive and visually-guided pursuit eye movements

    Experimental Brain Research 86:311–323.

    https://doi.org/10.1007/BF00228954

    • PubMed
    • Google Scholar
27. 1. Keller EL

    (1974) Participation of medial pontine reticular formation in eye movement generation in monkey

    Journal of Neurophysiology 37:316–332.

    https://doi.org/10.1152/jn.1974.37.2.316

    • PubMed
    • Google Scholar
28. 1. Kobak D
    2. Brendel W
    3. Constantinidis C
    4. Feierstein CE
    5. Kepecs A
    6. Mainen ZF
    7. Qi XL
    8. Romo R
    9. Uchida N
    10. Machens CK

    (2016) Demixed principal component analysis of neural population data

    eLife 5:e10989.

    https://doi.org/10.7554/eLife.10989

    • PubMed
    • Google Scholar
29. 1. Kodaka Y
    2. Kawano K

    (2003) Preparatory modulation of the gain of visuo-motor transmission for smooth pursuit in monkeys

    Experimental Brain Research 149:391–394.

    https://doi.org/10.1007/s00221-003-1375-y

    • PubMed
    • Google Scholar
30. 1. Komatsu H
    2. Wurtz RH

    (1989) Modulation of pursuit eye movements by stimulation of cortical Areas MT and MST

    Journal of Neurophysiology 62:31–47.

    https://doi.org/10.1152/jn.1989.62.1.31

    • PubMed
    • Google Scholar
31. 1. Kowler E
    2. Rubinstein JF
    3. Santos EM
    4. Wang J

    (2019) Predictive smooth pursuit eye movements

    Annual Review of Vision Science 5:223–246.

    https://doi.org/10.1146/annurev-vision-091718-014901

    • PubMed
    • Google Scholar
32. 1. Kuypers HG
    2. Brinkman J

    (1970) Precentral projections to different parts of the spinal intermediate zone in therhesus monkey

    Brain Research 24:29–48.

    https://doi.org/10.1016/0006-8993(70)90272-6

    • PubMed
    • Google Scholar
33. 1. Langer TP
    2. Kaneko CR

    (1990) Brainstem afferents to the oculomotor omnipause neurons in monkey

    The Journal of Comparative Neurology 295:413–427.

    https://doi.org/10.1002/cne.902950306

    • PubMed
    • Google Scholar
34. 1. Lisberger SG

    (2010) Visual guidance of smooth-pursuit eye movements: sensation, action, and what happens in between

    Neuron 66:477–491.

    https://doi.org/10.1016/j.neuron.2010.03.027

    • PubMed
    • Google Scholar
35. 1. Luschei ES
    2. Fuchs AF

    (1972) Activity of brain stem neurons during eye movements of alert monkeys

    Journal of Neurophysiology 35:445–461.

    https://doi.org/10.1152/jn.1972.35.4.445

    • PubMed
    • Google Scholar
36. 1. MacAvoy MG
    2. Gottlieb JP
    3. Bruce CJ

    (1991) Smooth-pursuit eye movement representation in the primate frontal eye field

    Cerebral Cortex 1:95–102.

    https://doi.org/10.1093/cercor/1.1.95

    • PubMed
    • Google Scholar
37. 1. Mahaffy S
    2. Krauzlis RJ

    (2011) Neural activity in the frontal pursuit area does not underlie pursuit target selection

    Vision Research 51:853–866.

    https://doi.org/10.1016/j.visres.2010.10.010

    • PubMed
    • Google Scholar
38. 1. Mazzoni P
    2. Bracewell RM
    3. Barash S
    4. Andersen RA

    (1996) Motor intention activity in the macaque's lateral intraparietal area. I. Dissociation of motor plan from sensory memory

    Journal of Neurophysiology 76:1439–1456.

    https://doi.org/10.1152/jn.1996.76.3.1439

    • PubMed
    • Google Scholar
39. 1. Messier J
    2. Kalaska JF

    (2000) Covariation of primate dorsal premotor cell activity with direction and amplitude during a memorized-delay reaching task

    Journal of Neurophysiology 84:152–165.

    https://doi.org/10.1152/jn.2000.84.1.152

    • PubMed
    • Google Scholar
40. 1. Missal M
    2. Heinen SJ

    (2001) Facilitation of smooth pursuit initiation by electrical stimulation in the supplementary eye fields

    Journal of Neurophysiology 86:2413–2425.

    https://doi.org/10.1152/jn.2001.86.5.2413

    • PubMed
    • Google Scholar
41. 1. Missal M
    2. Keller EL

    (2002) Common inhibitory mechanism for saccades and smooth-pursuit eye movements

    Journal of Neurophysiology 88:1880–1892.

    https://doi.org/10.1152/jn.2002.88.4.1880

    • PubMed
    • Google Scholar
42. 1. Munoz DP
    2. Wurtz RH

    (1995) Saccade-related activity in monkey superior colliculus. I. characteristics of burst and buildup cells

    Journal of Neurophysiology 73:2313–2333.

    https://doi.org/10.1152/jn.1995.73.6.2313

    • PubMed
    • Google Scholar
43. 1. Mustari MJ
    2. Ono S
    3. Das VE

    (2009) Signal processing and distribution in cortical-brainstem pathways for smooth pursuit eye movements

    Annals of the New York Academy of Sciences 1164:147–154.

    https://doi.org/10.1111/j.1749-6632.2009.03859.x

    • PubMed
    • Google Scholar
44. 1. Newsome WT
    2. Wurtz RH
    3. Komatsu H

    (1988) Relation of cortical Areas MT and MST to pursuit eye movements. II. differentiation of retinal from extraretinal inputs

    Journal of Neurophysiology 60:604–620.

    https://doi.org/10.1152/jn.1988.60.2.604

    • PubMed
    • Google Scholar
45. 1. Nuding U
    2. Kalla R
    3. Muggleton NG
    4. Büttner U
    5. Walsh V
    6. Glasauer S

    (2009) TMS evidence for smooth pursuit gain control by the frontal eye fields

    Cerebral Cortex 19:1144–1150.

    https://doi.org/10.1093/cercor/bhn162

    • PubMed
    • Google Scholar
46. 1. Perich MG
    2. Gallego JA
    3. Miller LE

    (2018) A neural population mechanism for rapid learning

    Neuron 100:964–976.

    https://doi.org/10.1016/j.neuron.2018.09.030

    • PubMed
    • Google Scholar
47. 1. Prut Y
    2. Fetz EE

    (1999) Primate spinal interneurons show pre-movement instructed delay activity

    Nature 401:590–594.

    https://doi.org/10.1038/44145

    • PubMed
    • Google Scholar
48. 1. Rashbass C

    (1961) The relationship between saccadic and smooth tracking eye movements

    The Journal of Physiology 159:326–338.

    https://doi.org/10.1113/jphysiol.1961.sp006811

    • PubMed
    • Google Scholar
49. 1. Robinson DA

    (1973) Models of the saccadic eye movement control system

    Kybernetik 14:71–83.

    https://doi.org/10.1007/BF00288906

    • Google Scholar
50. 1. Schoppik D
    2. Nagel KI
    3. Lisberger SG

    (2008) Cortical mechanisms of smooth eye movements revealed by dynamic covariations of neural and behavioral responses

    Neuron 58:248–260.

    https://doi.org/10.1016/j.neuron.2008.02.015

    • PubMed
    • Google Scholar
51. 1. Schwartz JD
    2. Lisberger SG

    (1994) Initial tracking conditions modulate the gain of visuo-motor transmission for smooth pursuit eye movements in monkeys

    Visual Neuroscience 11:411–424.

    https://doi.org/10.1017/S0952523800002352

    • PubMed
    • Google Scholar
52. 1. Shadlen MN
    2. Newsome WT

    (2001) Neural basis of a perceptual decision in the parietal cortex (area LIP) of the rhesus monkey

    Journal of Neurophysiology 86:1916–1936.

    https://doi.org/10.1152/jn.2001.86.4.1916

    • PubMed
    • Google Scholar
53. 1. Shi D
    2. Friedman HR
    3. Bruce CJ

    (1998) Deficits in smooth-pursuit eye movements after muscimol inactivation within the primate's frontal eye field

    Journal of Neurophysiology 80:458–464.

    https://doi.org/10.1152/jn.1998.80.1.458

    • PubMed
    • Google Scholar
54. 1. Stavisky SD
    2. Kao JC
    3. Ryu SI
    4. Shenoy KV

    (2017) Motor cortical visuomotor feedback activity is initially isolated from downstream targets in Output-Null neural state space dimensions

    Neuron 95:195–208.

    https://doi.org/10.1016/j.neuron.2017.05.023

    • PubMed
    • Google Scholar
55. 1. Tabata H
    2. Miura K
    3. Taki M
    4. Matsuura K
    5. Kawano K

    (2006) Preparatory gain modulation of visuomotor transmission for smooth pursuit eye movements in monkeys

    Journal of Neurophysiology 96:3051–3063.

    https://doi.org/10.1152/jn.00412.2006

    • PubMed
    • Google Scholar
56. 1. Tanaka M
    2. Fukushima K

    (1998) Neuronal responses related to smooth pursuit eye movements in the periarcuate cortical area of monkeys

    Journal of Neurophysiology 80:28–47.

    https://doi.org/10.1152/jn.1998.80.1.28

    • PubMed
    • Google Scholar
57. 1. Tanaka M
    2. Lisberger SG

    (2001) Regulation of the gain of visually guided smooth-pursuit eye movements by frontal cortex

    Nature 409:191–194.

    https://doi.org/10.1038/35051582

    • PubMed
    • Google Scholar
58. 1. Tanaka M
    2. Lisberger SG

    (2002a) Role of arcuate frontal cortex of monkeys in smooth pursuit eye movements. I. basic response properties to retinal image motion and position

    Journal of Neurophysiology 87:2684–2699.

    https://doi.org/10.1152/jn.2002.87.6.2684

    • PubMed
    • Google Scholar
59. 1. Tanaka M
    2. Lisberger SG

    (2002b) Role of arcuate frontal cortex of monkeys in smooth pursuit eye movements. II. relation to vector averaging pursuit

    Journal of Neurophysiology 87:2700–2714.

    https://doi.org/10.1152/jn.2002.87.6.2700

    • PubMed
    • Google Scholar
60. 1. Tanji J
    2. Evarts EV

    (1976) Anticipatory activity of motor cortex neurons in relation to direction of an intended movement

    Journal of Neurophysiology 39:1062–1068.

    https://doi.org/10.1152/jn.1976.39.5.1062

    • PubMed
    • Google Scholar
61. 1. Weinrich M
    2. Wise SP

    (1982) The premotor cortex of the monkey

    The Journal of Neuroscience 2:1329–1345.

    https://doi.org/10.1523/JNEUROSCI.02-09-01329.1982

    • PubMed
    • Google Scholar
62. 1. Wurtz RH
    2. Goldberg ME

    (1972) Activity of superior colliculus in behaving monkey. 3. cells discharging before eye movements

    Journal of Neurophysiology 35:575–586.

    https://doi.org/10.1152/jn.1972.35.4.575

    • PubMed
    • Google Scholar

Acceptance summary:

The study investigates how activity in a specific region of the frontal eye field changes when smooth pursuit eye movements switch from preparation to execution. They propose a novel framework based on visuomotor gain to account for the behavioural and neural data collected from monkeys, different from the movement potent framework that has been previously proposed for the skeletomotor system.

Decision letter after peer review:

Thank you for submitting your article entitled "Mechanisms that allow cortical preparatory activity without inappropriate movement" for peer review at eLife. Your article is being evaluated by three peer reviewers, and the evaluation is being overseen by a Reviewing Editor and Joshua Gold as the Senior Editor.

After consultation with each other and with the Reviewing Editor and Senior Editor, the reviewers have identified a list of essential revisions, involving extensive new analyses, which are listed below.

The present study aims to understand whether pseudo-population activity of FEFsem neurons is better explained by a visual-motor gain hypothesis or by the movement-potent/null space framework for the skeletomotor system. Using a brief pulse of visual motion during the preparatory period to produce transient eye motion during fixation, the authors found that the size of eye motion is modulated by factors such as the timing of pulse within the fixation period, direction of pulse motion, and expectation of target speed. This result was interpreted as matching the predictions of the visuomotor gain hypothesis. The authors further found that: 1) an optimal linear model, parameterized by weight estimates from one smooth pursuit task with 8 directions, was used on the same FEFsem neuron population during the preparatory period, and the model predictions appeared to match the actual eye velocity during fixation when there wasn't any; and 2) the principal components (PCs) of the population response measured during preparatory and movement initiation phases were separated by about 65 deg, as opposed to 90 deg as predicted by the movement-null space framework. These results were interpreted as evidence against the movement-null space framework.

All three reviewers (including the reviewing editor) find this paper interesting, well-written, and timely for the field. However, they converged on a common set of concerns about the data analysis.

1) The evidence against the movement-null space frame is indirect. For example, the approach is to find the preparatory subspace in one task and look at what happens in another task. Extensive additional analyses are needed to make the current work comparable to the studies (e.g., Kaufman et al.).

2) Some data analyses are confusing, such as artificially rotating the tuning of some neurons to build a neural space that sometimes includes the same neuron on different axes. These analyses should be either simplified or more clearly explained and justified.

3) The PCA analysis should be improved; e.g., using dPCA.

4) It would be meaningful and necessary to applied subspace analysis to the data from the experiment that provided the behavioral evidence (e.g., Figure 2).

Reviewer #1:

In this manuscript Darlington and Lisberger address the question how preparatory activity in the smooth pursuit eye movement system is prevented from actually driving eye movements. Different solutions have been proposed for different motor systems. The saccadic eye movement system has a gate (omnipause neurons) preventing preparatory activity from activating burst neurons. An analysis of primary motor and premotor activity in the arm movement system has shown that preparatory population activity avoids state space dimensions that result in driving motor output. For the smooth pursuit eye movement system, the authors propose that preparatory activity in the frontal eye field (FEF) controls the gain of a sensorimotor transformation process that is actually driven by visual motion information. An analysis of the FEF population activity indicated that the preparatory activity does not completely avoid the dimensions related to motor output, such that the preparatory activity would be expected to cause eye movements if the neural activity in FEF were responsible for driving smooth pursuit.

1) One of the key elements supporting the authors' claims is a comparison of the time course of preparatory activity in FEF with the time course of the behavioral effect of brief motion pulses. Ideally, these data would have been collected in a combined experiment. For this manuscript they were collected in separate experiments. This is not necessarily problematic as they were collected from the same monkeys. However, both time courses are likely affected by the animals' temporal expectations about upcoming eye movements. The timing of the experiments (distribution of time intervals between fixation onset and onset of the continuous motion cue driving pursuit, not the pulses) should therefore have been identical to be able to make the assumption that the animals' expectation was probably similar. The manuscript is not explicit about whether this was the case. Was it?

Other than that, I don't have any major issues. The data analysis is solid, and the manuscript is very clearly written.

Reviewer #2:

Darlington and Lisberger present a well-written and interesting study that explores how activity in a specific region of the frontal eye field (FEFsem) transitions between preparation and execution of smooth pursuit eye movements. The overall objective is to understand whether pseudo-population activity of FEFsem neurons is better explained by the visual-motor gain hypothesis, originally developed by Lisberger and colleagues, or by the movement-potent/null space framework championed recently by Shenoy and colleagues for the skeletomotor system. The authors provide three results that, in their opinion, support the visual-motor gain hypothesis and refute the movement-null space perspective. I will summarize the three findings and then evaluate the authors' interpretations.

1) Injecting a brief pulse of visual motion during the preparatory period produces transient eye motion during fixation. The size of eye motion depends on several factors (when during the fixation period, direction of pulse motion, and expectation of target speed). The authors view the results as consistent with and build logically on the visual-motor gain hypothesis as a mechanism for movement initiation, particularly smooth pursuit. In my view, their interpretation is reasonable.

2) As a test of the movement-potent space hypothesis, the authors estimate the contribution of each neuron/condition to the initiation of smooth pursuit during one task condition (8 directions randomly interleaved). They then use these weights on the preparatory activity of the same population during a different task (repeat trials of same direction) and find that the model predicts changes in eye velocity (during fixation) when there isn't any. This finding is interpreted as evidence against the movement-potent space framework.

In my view, the analysis is not rigorous enough to reach this conclusion. Is it safe to assume, as the authors seem to do, that the population response for movement initiation during the two tasks is the same? How well do the weights for the 8D task predict pursuit initiation for the 1D task? Also, how well do weights obtained from 1D task during pursuit predict eye velocity during preparation in the same 1D task? These factors must be addressed before appreciating the importance of the result emphasized in the manuscript. Also see dPCA comment below.

3) In a related analysis, the authors evaluated the principal components (PCs) of the population response measured during preparatory and movement initiation phases. They found that the 1st PC for each period accounts for ~70% of the variance and that the mean angle between these PCs is about 65 deg, which is not equal to the 90 deg required if preparatory period activity is in a movement-null space. In addition, they identified a subpopulation of neurons that accounts for the observed change in angle between the PCs.

Given that the angle of separation between the primary preparatory and movement PCs is not 90 deg, the authors de-emphasize the potential importance of the movement potent/null space framework and instead view favorably the visual-motor gain hypothesis. However, they do state, "two mechanisms for preventing movements might be better than one". My view on their data is that the population FEFsem definitely resides in different subspaces during preparatory and movement initiation periods. In other words, if we relax a strict interpretation of orthogonality of movement and null subspaces, then these data are very much in line with potent/null space framework. This very much contrasts the take-home message the authors deliver.

Moreover, it is not clear that PC analysis is the best/correct analysis to use for the current data. The authors should consider using demixed principal component analysis (dPCA; https://elifesciences.org/articles/10989; https://www.eneuro.org/content/3/4/ENEURO.0085-16.2016.long). An interesting outcome of previous studies using dPCA is that the primary PCA is generally condition independent and could reflect the large overlap in angle between the PCs. A separation is more appreciable with secondary and tertiary PCs. The authors should pursue this angle of inquiry.

Finally, there is a glaring omission in the analysis. The first part of the paper shows a robust behavioral result, in which a brief visual motion perturbation produces an eye movement during a period typically associated with movement preparation. What subspace is spanned by FEFsem population during the transient eye movement response? Does the activity remain aligned with preparatory PCs (evidence against functional importance of movement-null space) or does it shift to movement PCs (evidence to support the movement-potent hypothesis)? It seems to me that this analysis is crucial to understand population activity control of smooth pursuit by FEFsem.

In summary, a study of how population activity relates to movement preparation and generation is timely. Tests of whether new hypotheses developed in skeletomotor systems generalize to the oculomotor system is very much needed, and this study takes a step in the direction. However, the analysis isn't rigorous enough to reach a reliable conclusion.

Reviewer #3:

Darlington and Lisberger present a previously undescribed "mechanism" by which the smooth eye movement (SEM) region of FEF could prepare pursuit movements without causing them -by dialing up the visuomotor gain. Moreover, they provide evidence in the SEM system against the "null subspace" hypothesis of preparation of arm movements. I find many aspects of the manuscript interesting but I have some concerns about their analytical approach

Comments:

1) Although I think that the potential differences in control of reaching movements and pursuits are interesting, I have concerns about how the authors tested the "null subspace" hypothesis. Overall, I think that the authors should replicate the methods proposed by Kaufman et al., rather than introduce a number of modifications whose implications are hard to grasp -at least for me. In more detail:

1a) All subspace analyses putatively capture activity that reflects some aspect of the underlying circuitry as well as the task (e.g, related to inputs to the neural population or outputs). However, the authors find the null subspace in a lower-dimensional version of the task that they probe (1-targets vs. 8-targets) that is also -I think- cued differently. My suggestion is that they do the analysis using the full 8-target data and considering the preparatory and pursuit epochs as Kaufman et al. did.

1b) Studies like Kaufman's assume that the relevant "neural information" is captured by activity patterns shared across the neural population; these patterns are probed by analyzing a state space in which axis is the activity of one recorded neuron. The authors adopted an approach that puzzled me: having multiple axes that reflect the same neuron after rotating its activity. I'd again suggest them to adopt the simpler approach of defining one axes per neuron, discarding neurons that were recorded multiple times during different conditions (at least I have trouble foreseeing all potential implications of their manipulations)

1c) I am again a bit confused about the potential effect of the authors rotating the tuning curves of each neuron in the analysis. Although I agree that there may be FEF neurons with these properties, I would avoid it because it implies assuming that the absolute coordinates of the target doesn't matter, or at least show that it doesn't influence the main result.

1d) Did the authors include all the recorded neurons in this analysis or only neurons that were significantly tuned to the task? Although studies in the motor cortices suggest that the population dynamics sampled from sufficiently large neural populations are the same irrespective of the sampled neurons (Trautman et al., 2018), I worry this may bias their results.

1e) Because of these concerns, I believe that the authors should repeat Figure 8 after calculating the preparatory and pursuit PCs as in Kaufman et al., 2014.

1f) Minor: I assume that the authors only compare the first pursuit PC and the first preparatory PC because they each capture a large fraction of the total neural variance. For a full assessment, I'd suggest an analysis like the one in Figure 4 of Elsayed et al., 2016, or the principal angle analysis used in Gallego et al., 2018.

2) The central idea of "gain control" is interesting, but I think the authors should pursue it further. For example, can single trial eye speed be predicted from neural activity? (I'd do this analysis for the data in Figure 2 and 3). Also, how do the authors think this gain leads to a transition from preparation to movement, by exceeding a threshold? I think some analyses and discussion in this regard are missing.

3) In the Results section the authors make the proposal that there may be "non-specific preparatory enhancement of visuomotor gain". I think they could address this directly by using demixed principal component analysis (dPCA) (Kobak et al., 2016). dPCA effectively performs semi-supervised dimensionality reduction and can be used to isolate population activity patterns that only depend on time (these would be the component the authors refer to) and components that depend on the direction of the movement (see, e.g., Kaufman et al., 2016, Gallego et al., 2018 for examples in motor cortex). Examining both the dynamics of these components (e.g., when do putative time-related and target-related components start ramping up?) and the weights (e.g., are there subpopulations of neurons?) could be interesting. Note that there's freely available code by Kobak et al.

4) Figure 2 shows strong effects, but it would be nice to see the data distributions, or at least box plot versions of (b) and (c). As to (d), would the authors see the same trend if they plotted single trial values or a 2D histogram? This figure focuses in neurons that had "positive preparatory firing rate modulation". What was their percentage among all recorded neurons? And among all modulated neurons? And what happens with these neurons? Same for the increase and decrease cells in Figure 3.

References:

GF Elsayed, AH Lara, MT Kaufman, MM Churchland and JP Cunningham. Reorganization between preparatory and movement population responses in motor cortex. Nature Communications 2016

JA. Gallego, MG. Perich, SN. Naufel, C Ethier, SA Solla and LE Miller. Cortical population activity within a preserved neural manifold underlies multiple motor behaviors. Nature Communications 2018

MT Kaufman, JS Seely, D Sussillo, SI Ryu, KV Shenoy and MM Churchland. The Largest Response Component in the Motor Cortex Reflects Movement Timing but Not Movement Type. eNeuro 2016

D Kobak, W Brendel, C Constantinidis, CE Feierstein, A Kepecs, ZF Mainen, X-L Qi, R Romo, N Uchida, C Machens. Demixed principal component analysis of neural population data. eLife 2016

MG Perich, JA Gallego, LE Miller. A Neural Population Mechanism for Rapid Learning. Neuron 2018

EM Trautmann, SD Stavisky, S Lahiri, KC Ames, MT Kaufman, DJ O'Shea, S Vyas, X Sun, SI Ryu, S Ganguli, KV Shenoy. Accurate estimation of neural population dynamics without spike sorting. Neuron 2018

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Mechanisms that allow cortical preparatory activity without inappropriate movement" for further consideration by eLife. Your revised article has been evaluated by Joshua Gold (Senior Editor), a Reviewing Editor, and two of the original reviewers.

The manuscript has been improved substantially. The addition of new analyses as well as the justification for why other requested analyses are not needed were satisfactory. Overall, the revised version is well written and of strong impact. It pushes back a bit against the increasingly dogmatic view of movement-potent and movement-null subspaces.

We have one remaining major concern that should be addressed:

1) It would be interesting to a) recompute weights relating FEFsem activity to eye velocity separately for subpopulations 1 and 2, and (b) predict eye velocities for each. Is there a difference in the predictive power for the two subpopulations? And how well do the results align with the proposed role of each subset (including for different directions)? A new figure can probably be committed to the results.

Reviewer #1:

In this manuscript Darlington and Lisberger address the question how preparatory activity in the smooth pursuit eye movement system is prevented from actually driving eye movements. Different solutions have been proposed for different motor systems. The saccadic eye movement system has a gate (omnipause neurons) preventing preparatory activity from activating burst neurons. An analysis of primary motor and premotor activity in the arm movement system has shown that preparatory population activity avoids state space dimensions that result in driving motor output. For the smooth pursuit eye movement system, the authors propose that preparatory activity in the frontal eye field (FEF) controls the gain of a sensorimotor transformation process that is actually driven by visual motion information. An analysis of the FEF population activity indicated that the preparatory activity does not completely avoid the dimensions related to motor output, such that the preparatory activity would be expected to cause eye movements if the neural activity in FEF were responsible for driving smooth pursuit.

1) One of the key elements supporting the authors' claims is a comparison of the time course of preparatory activity in FEF with the time course of the behavioral effect of brief motion pulses. Ideally, these data would have been collected in a combined experiment. For this manuscript they were collected in separate experiments. This is not necessarily problematic as they were collected from the same monkeys. However, both time courses are likely affected by the animals' temporal expectations about upcoming eye movements. The timing of the experiments (distribution of time intervals between fixation onset and onset of the continuous motion cue driving pursuit, not the pulses) should therefore have been identical to be able to make the assumption that the animals' expectation was probably similar. The manuscript is not explicit about whether this was the case. Was it?

We agree that this is a very important issue. Despite the current ordering of figures, the behavioral experiments were actually a follow-up to the neural analyses we performed on our FEFSEM data. We matched all experimental conditions, including timing, in order to maximize the validity of our comparisons between the neurophysiology and behavioral data sets. We have added a short paragraph in the relevant place in the Results to be more explicit about this.

Other than that, I don't have any major issues. The data analysis is solid, and the manuscript is very clearly written.

We’d like to thank the reviewer for the positive feedback. We are glad that the reviewer enjoyed our manuscript.

Reviewer #2:

Darlington and Lisberger present a well-written and interesting study that explores how activity in a specific region of the frontal eye field (FEFsem) transitions between preparation and execution of smooth pursuit eye movements. The overall objective is to understand whether pseudo-population activity of FEFsem neurons is better explained by the visual-motor gain hypothesis, originally developed by Lisberger and colleagues, or by the movement-potent/null space framework championed recently by Shenoy and colleagues for the skeletomotor system. The authors provide three results that, in their opinion, support the visual-motor gain hypothesis and refute the movement-null space perspective. I will summarize the three findings and then evaluate the authors' interpretations.

1) Injecting a brief pulse of visual motion during the preparatory period produces transient eye motion during fixation. The size of eye motion depends on several factors (when during the fixation period, direction of pulse motion, and expectation of target speed). The authors view the results as consistent with and build logically on the visual-motor gain hypothesis as a mechanism for movement initiation, particularly smooth pursuit. In my view, their interpretation is reasonable.

2) As a test of the movement-potent space hypothesis, the authors estimate the contribution of each neuron/condition to the initiation of smooth pursuit during one task condition (8 directions randomly interleaved). They then use these weights on the preparatory activity of the same population during a different task (repeat trials of same direction) and find that the model predicts changes in eye velocity (during fixation) when there isn't any. This finding is interpreted as evidence against the movement-potent space framework.

In my view, the analysis is not rigorous enough to reach this conclusion. Is it safe to assume, as the authors seem to do, that the population response for movement initiation during the two tasks is the same? How well do the weights for the 8D task predict pursuit initiation for the 1D task? Also, how well do weights obtained from 1D task during pursuit predict eye velocity during preparation in the same 1D task? These factors must be addressed before appreciating the importance of the result emphasized in the manuscript. Also see dPCA comment below.

We’d like to thank the reviewer for the insightful questions and we agree that it is important that they be answered. In Figure 5D, which was part of the original paper, we show that the weights calculated in the 8D task do an excellent job of predicting actual horizontal eye velocity for the 1D task. These same weights accurately predict 0 vertical eye velocity (i.e. accurately predict direction) for the 1D task; because this is a ridiculous looking figure with all points plotting at (0,0), we state this result explicitly in the text. We have now also included a supplementary analysis shown in Figure 6—figure supplement 3 to confirm that we obtained all the same results when the fitting and predicting were both done on the 1D task.

3) In a related analysis, the authors evaluated the principal components (PCs) of the population response measured during preparatory and movement initiation phases. They found that the 1st PC for each period accounts for ~70% of the variance and that the mean angle between these PCs is about 65 deg, which is not equal to the 90 deg required if preparatory period activity is in a movement-null space. In addition, they identified a subpopulation of neurons that accounts for the observed change in angle between the PCs.

Given that the angle of separation between the primary preparatory and movement PCs is not 90 deg, the authors de-emphasize the potential importance of the movement potent/null space framework and instead view favorably the visual-motor gain hypothesis. However, they do state, "two mechanisms for preventing movements might be better than one". My view on their data is that the population FEFsem definitely resides in different subspaces during preparatory and movement initiation periods. In other words, if we relax a strict interpretation of orthogonality of movement and null subspaces, then these data are very much in line with potent/null space framework. This very much contrasts the take-home message the authors deliver.

We’d like to thank the reviewer for this feedback, which helped us to recognize a deficiency in our presentation in the original version of the paper. We have rewritten parts of the Results and Discussion to be transparent that while not orthogonal, there is clearly some reorganization between preparation and movement. We also added text to explain that the full set of data that we think argues against a simple and traditional “null-space” explanation for our findings. First, preparatory activity does evolve in a way that predicts movement, which is the strongest piece of evidence against the null/potent framework in the pursuit system. Second, we outline a broader (and in our view more compelling) function for this reorganization beyond simply preventing movement during preparation. Through an analysis that isolates the part of our population that is driving the partial reorganization, cells whose preferred direction was generally opposite the direction of the 1D experiment (that seemingly have a strong “prior” for that direction based on their preparatory responses), we connect this feature of the full population response to our behavioral finding of a directionally non-specific enhancement of visual motor gain (Figure 4A, orthogonal/opposite). We suggest that a more likely function of this reorganization is to allow the pursuit system to partially prepare for less likely directions of visual motion, so that it is never caught completely flat-footed in the event of unexpected target motion.

Moreover, it is not clear that PC analysis is the best/correct analysis to use for the current data. The authors should consider using demixed principal component analysis (dPCA; https://elifesciences.org/articles/10989; https://www.eneuro.org/content/3/4/ENEURO.0085-16.2016.long). An interesting outcome of previous studies using dPCA is that the primary PCA is generally condition independent and could reflect the large overlap in angle between the PCs. A separation is more appreciable with secondary and tertiary PCs. The authors should pursue this angle of inquiry.

We considered dPCA and read the paper cited by the reviewer, but decided in the end that our unbiased approach was actually more rigorous. We wanted to use an unbiased estimator of preparatory and pursuit dimensions to produce an unbiased result. Unless we have misunderstood the reviewer, we disagree that it would be a move rigorous approach to show that preparatory and pursuit secondary and tertiary PCs are different based on the results of semi-supervised dimensionality reduction, which defines preparation and pursuit as being different conditions.

Finally, there is a glaring omission in the analysis. The first part of the paper shows a robust behavioral result, in which a brief visual motion perturbation produces an eye movement during a period typically associated with movement preparation. What subspace is spanned by FEFsem population during the transient eye movement response? Does the activity remain aligned with preparatory PCs (evidence against functional importance of movement-null space) or does it shift to movement PCs (evidence to support the movement-potent hypothesis)? It seems to me that this analysis is crucial to understand population activity control of smooth pursuit by FEFsem.

Unfortunately, we do not have access to the data requested by the reviewer. The visual motion pulse experiments presented in Figures 2-4 were behavior-only experiments done as a direct follow-up to test our interpretation of the neural analyses of Figures 5-8.

First, we want to argue that this is not a glaring omission. The responses to a pulse of target motion at different times during fixation do not differ conceptually from the response to the onset of target motion at the end of the prescribed fixation period. Both initiate eye movement, both are driven by visual motion, and both represent neural correlates of the eye movement behavioral response. There is no reason to think that they would differ from the pursuit-related responses and the population activity during a probe should, by all accounts, align with the movement PCs (and only be partially reorganized from preparatory PCs). To collect these data would be a ton of work only to confirm a very solid prediction.

Second, we want to make an offer. A lab member who is not an author in this paper has collected neurophysiological data in FEFSEM during visual motion pulse experiments (n=51 neurons in one monkey). Even though there are some minor differences in the experimental design, we have analyzed those data and found a positive correlation (r=0.62) between preparatory modulation of firing rate and modulation of activity during the pulses (see Author response image 1). Thus, population modulation in response to the pulses is in the same direction as modulation during preparation. This is consistent with the relationship that we have shown between preparatory- and pursuit-related firing rates (Figure 1D) and weakly supports our expectation from doing this experiment correctly. We would be happy to include this as a supplementary figure and add the other lab member as an author on the paper if the reviewer deems that it materially adds to the story, even though we do not think it is a strong addition.

Author response image 1

Download asset Open asset

In summary, a study of how population activity relates to movement preparation and generation is timely. Tests of whether new hypotheses developed in skeletomotor systems generalize to the oculomotor system is very much needed, and this study takes a step in the direction. However, the analysis isn't rigorous enough to reach a reliable conclusion.

We appreciate the positive part of this comment. We have done the best we can to improve the rigor along the lines recommended by the reviewer, but we also respectfully disagree with one of the major comments concerning the use of dPCA. We agree that dPCA would be more “modern”, and have added dPCA in another portion of the manuscript. However, we think the analysis we have done in comparing preparatory and pursuit PCs is closer to unbiased and that this is a major virtue.

Reviewer #3:

Darlington and Lisberger present a previously undescribed "mechanism" by which the smooth eye movement (SEM) region of FEF could prepare pursuit movements without causing them -by dialing up the visuomotor gain. Moreover, they provide evidence in the SEM system against the "null subspace" hypothesis of preparation of arm movements. I find many aspects of the manuscript interesting but I have some concerns about their analytical approach.

Comments:

1) Although I think that the potential differences in control of reaching movements and pursuits are interesting, I have concerns about how the authors tested the "null subspace" hypothesis. Overall, I think that the authors should replicate the methods proposed by Kaufman et al., rather than introduce a number of modifications whose implications are hard to grasp -at least for me.

We’d like to thank the reviewer for this feedback and agree that we did not justify/clarify our use of these modifications. The primary reason we modified our analytical approach is because there are key experimental differences between the way that we collected our data and the way in which the Kaufman et al. data was collected. We have added text to clarify. Furthermore, we have taken two major steps to remedy this concern. First, we have included control analyses to show that our conclusions are not due to the modifications. Second, we have included as Figure 6—figure supplement 3 a supplementary analysis that is more congruent with what was done in the Kaufman et al. paper.

In more detail:

1a) All subspace analyses putatively capture activity that reflects some aspect of the underlying circuitry as well as the task (e.g, related to inputs to the neural population or outputs). However, the authors find the null subspace in a lower-dimensional version of the task that they probe (1-targets vs. 8-targets) that is also -I think- cued differently. My suggestion is that they do the analysis using the full 8-target data and considering the preparatory and pursuit epochs as Kaufman et al. did.

The only difference between the 1-direction and 8-direction tasks is the preponderance of target directions that occurs. There were no explicit cues in either task indicating the direction of the upcoming target motion. Therefore, accurate preparation is only possible during the 1-direction task where the monkey is able to anticipate the upcoming target motion. We decided to use the 8-direction task to perform the mapping of pursuit-related FEFSEM activity to eye movement because it provides a richer assessment of eye velocity. We have written more clearly about this issue, in a way that we hope resolves the reviewer’s concerns.

1b) Studies like Kaufman's assume that the relevant "neural information" is captured by activity patterns shared across the neural population; these patterns are probed by analyzing a state space in which axis is the activity of one recorded neuron. The authors adopted an approach that puzzled me: having multiple axes that reflect the same neuron after rotating its activity. I'd again suggest them to adopt the simpler approach of defining one axes per neuron, discarding neurons that were recorded multiple times during different conditions (at least I have trouble foreseeing all potential implications of their manipulations)

1c) I am again a bit confused about the potential effect of the authors rotating the tuning curves of each neuron in the analysis. Although I agree that there may be FEF neurons with these properties, I would avoid it because it implies assuming that the absolute coordinates of the target doesn't matter, or at least show that it doesn't influence the main result.

We can understand how readers might be confused by some of the choices we used in our data analysis and we apologize for not being clearer in our explanations and for not including some obvious controls. The goal was to create as much of a population response as we could, without excluding data points or neurons and without biasing our data collection for the 1D experiment towards any part of the direction tuning curve of our recorded neurons (i.e. only running the 1D experiment in the preferred direction of our neurons). Therefore, we ran multiple single direction experiments (where preparation could effectively occur) for most neurons in different directions relative to their preferred direction. We then chose to rotate the direction tuning curves in a manner that made it seem like all of the 1D experiments were run in the same direction (0 degrees, rightward). This allowed us to construct a full pseudo-population preparatory response for which the full range of differences between the expected direction of the upcoming visual motion and the preferred direction of the cell is represented.

We have corrected our prior errors. Now, we include a control analysis where each neuron contributes a single data point and we show that our conclusions are unchanged (Figure 6—figure supplement 2). Furthermore, Figure 6—figure supplement 3 show that we obtain the same results when the analysis used only data from the 1D task and involved no rotation of neural data.

1d) Did the authors include all the recorded neurons in this analysis or only neurons that were significantly tuned to the task? Although studies in the motor cortices suggest that the population dynamics sampled from sufficiently large neural populations are the same irrespective of the sampled neurons (Trautman et al., 2018), I worry this may bias their results.

We included all recorded neurons in this analysis. We are now clear about this in the text.

1e) Because of these concerns, I believe that the authors should repeat Figure 8 after calculating the preparatory and pursuit PCs as in Kaufman et al., 2014.

We have now included as Figure 6—figure supplement 3 a supplementary analysis that is more congruent with that used in Kaufman et al., 2014 in that it uses data from only one task, computes a weight matrix between pursuit PCs and eye movement, and tests the projections of preparatory PCs onto the row and null spaces of the weight matrix. The results and conclusions are unchanged.

1f) Minor: I assume that the authors only compare the first pursuit PC and the first preparatory PC because they each capture a large fraction of the total neural variance. For a full assessment, I'd suggest an analysis like the one in Figure 4 of Elsayed et al., 2016, or the principal angle analysis used in Gallego et al., 2018.

We chose only the first PCs because they each captured ~70% of the variance, the same value used to choose the number of PCs used in Elsayed et al., 2016. We now state this clearly in the text.

2) The central idea of "gain control" is interesting, but I think the authors should pursue it further. For example, can single trial eye speed be predicted from neural activity? (I'd do this analysis for the data in Figure 2 and 3). Also, how do the authors think this gain leads to a transition from preparation to movement, by exceeding a threshold? I think some analyses and discussion in this regard are missing.

We have added some text to the Discussion to pursue our ideas about “gain control” further. A previous publication from our laboratory showed that single trial eye speed can be predicted from neural activity during the initiation of pursuit (Schoppik et al., 2008). Because Figures 2 and 3 show the results of behavioral experiments without neural recordings, we cannot address this cogent question with the data at hand. We would, however, point out that single-trial neuron-behavior correlations are more of a statement about neuron-neuron correlations than about any particular functional role and, so, we view that question as tangential to the main point of the present paper. In terms of what triggers movement, we have just published a paper that shows the absence of a threshold for triggering movement based on the signals in the FEFSEM (Lee et al., 2019). Instead, we think that the gain signal from FEFSEM poises the system for action, but movement is triggered only by visual motion signals that are subject to the gain modulation.

3) In the Results section the authors make the proposal that there may be "non-specific preparatory enhancement of visuomotor gain". I think they could address this directly by using demixed principal component analysis (dPCA) (Kobak et al., 2016). dPCA effectively performs semi-supervised dimensionality reduction and can be used to isolate population activity patterns that only depend on time (these would be the component the authors refer to) and components that depend on the direction of the movement (see, e.g., Kaufman et al., 2016, Gallego et al., 2018 for examples in motor cortex). Examining both the dynamics of these components (e.g., when do putative time-related and target-related components start ramping up?) and the weights (e.g., are there subpopulations of neurons?) could be interesting. Note that there's freely available code by Kobak et al.

We’d like to thank the reviewer for this excellent suggestion. We used dPCA on a subset of our FEFSEM population in which we were able to collect single-direction data both toward and away from the preferred direction of each neuron. The results are shown in Figure 4 and reveal a very nice correlate with the behavioral findings of Figure 4A, making an even stronger link between FEFSEM preparatory activity and the preparatory modulation of visual-motor gain.

4) Figure 2 shows strong effects, but it would be nice to see the data distributions, or at least box plot versions of (b) and (c). As to (d), would the authors see the same trend if they plotted single trial values or a 2D histogram? This figure focuses in neurons that had "positive preparatory firing rate modulation". What was their percentage among all recorded neurons? And among all modulated neurons? And what happens with these neurons? Same for the increase and decrease cells in Figure 3.

The source data for Figure 2 (a and b) panels are included with the submission. We have added information to the paper clarifying the percentage of neurons categorized as having positive versus negative preparatory firing rate modulation. The behavioral data for the pulses was collected separately (though under identical conditions) from the neurophysiology data. Therefore, it is not possible to do this analysis with single trials. However, our prior papers (Darlington et al., 2018; Lee et al., 2019) show that there is a positive relationship between fixation time and preparatory firing rate, and unpublished data show a correlation between preparatory firing rate and ensuing pursuit speed on a trial-by-trial basis. We would be willing to include the latter data in Figure 2, but it seems somewhat tangential to the main points of this paper.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

We have one remaining major concern that should be addressed:

1) It would be interesting to a) recompute weights relating FEFsem activity to eye velocity separately for subpopulations 1 and 2, and (b) predict eye velocities for each. Is there a difference in the predictive power for the two subpopulations? And how well do the results align with the proposed role of each subset (including for different directions)? A new figure can probably be committed to the results.

We performed the linear model analysis from Figures 5 and 6 separately on the two subpopulations. This analysis has been added to the manuscript as Figure 9. It reveals that both subpopulations perform roughly the same in terms of predicting the eye velocity during pursuit in the single-direction block. Furthermore, it shows that preparatory activity in subpopulation 1 evolves in a way that predicts movement in the direction of the upcoming visual motion and eye movement. However, preparatory activity from subpopulation 2 predicts no movement on average.

Author details

1. Timothy R Darlington

   Department of Neurobiology, Duke University School of Medicine, Durham, United States

   Contribution

   Conceptualization, Formal analysis, Investigation, Methodology, Writing - original draft, Writing - review and editing

   For correspondence

   timothy.r.darlington@gmail.com

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-5534-7552

2. Stephen G Lisberger

   Department of Neurobiology, Duke University School of Medicine, Durham, United States

   Contribution

   Conceptualization, Supervision, Funding acquisition, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7859-4361

• 1,981

  Page views

• 232

  Downloads

• 9

  Citations

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