Mechanisms that allow cortical preparatory activity without inappropriate movement

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.

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).

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.

Preparatory activity is ubiquitous in motor systems. In advance of all but the most short-latency and reflexive movements, preparatory activity appears at least in the primary motor cortex (Tanji and Evarts, 1976), the premotor cortex (Weinrich and Wise, 1982), the frontal eye fields for both saccades and pursuit (Bruce and Goldberg, 1985; Mahaffy and Krauzlis, 2011; Tanaka and Fukushima, 1998), the parietal cortex (Mazzoni et al., 1996; Shadlen and Newsome, 2001), the cerebellum (Gao et al., 2018), the superior colliculus (Dorris et al., 1997; Glimcher and Sparks, 1992; Munoz and Wurtz, 1995; Wurtz and Goldberg, 1972), and even in last-order interneurons in the spinal cord (Prut and Fetz, 1999). It seems clear that preparatory activity plays a crucial role in setting the state of the nervous system leading up to movements. Many of the neurons that show preparatory changes in firing rate seem to play key roles in selecting, initiating, and/or guiding movement. Yet, preparatory activity evolves without actually causing movement. Why?

Different mechanistic explanations, all supported by data, have been proposed for different motor systems. The earliest solution, proposed for saccadic eye movements, is a ‘gate’ or ‘switch’ (Keller, 1974). The neural circuit that implements the switch is in the brainstem (Cohen and Henn, 1972; Evinger et al., 1982; Luschei and Fuchs, 1972). It has known internal connectivity, documented by electrophysiology and anatomical tracing with horseradish peroxidase (Hikosaka et al., 1978; Langer and Kaneko, 1990). The connectivity leads to switching behavior in computational models and the output pathways from the brain’s switching circuit are appropriate to prevent saccades during preparation until it is time for the movement (Fuchs et al., 1985; Robinson, 1973).

A more recent solution, proposed for arm movements, suggests that the population of neurons in the motor cortex operates as a dynamical system (Churchland et al., 2010; Churchland et al., 2012). During the preparation of a movement, the population activity evolves within a set of dimensions that are ‘movement-null’ in the sense that population output is orthogonal to a later set of states that exists in ‘movement-potent’ dimensions (Elsayed et al., 2016; Kaufman et al., 2014). Analysis of population activity in the motor and pre-motor cortex supports the dynamical system explanation.

A third mechanism, based on our and others’ research in smooth pursuit eye movements, involves the use of motor cortical output signals not to drive movement, but instead to modulate the strength of sensory-motor transmission. In the pursuit system, several lines of evidence, most notably the effects of microstimulation, imply that the output from the FEF_SEM modulates the gain of visual-motor transmission (Nuding et al., 2009; Tanaka and Lisberger, 2001).

We imagine that all these mechanisms could be used in different blends by different motor systems. So, the challenge is to identify the mechanisms that prevent preparatory activity from causing movement in as many different movements as possible, explore the blend of mechanisms deployed by each motor system, and find the neural substrates of the different mechanisms.

In the present paper, we present evidence that the smooth pursuit eye movement system emphasizes gain-modulation as a mechanism to allow preparation to progress without causing movement. Even though preparatory activity in FEF_SEM is movement-potent in the sense that it shares dimensions with pursuit-related activity of the same population, preparatory activity fails to cause smooth eye movement because its role is to modulate the strength of movement-potent visual motion signals that are not present during preparation. Without the visual motion signal that it acts upon, gain can be dialed up without causing movement. The smooth eye movement output of the system will remain zero.

In contrast, our population analysis fails to support the dynamical systems explanation of orthogonal preparatory- and movement-related neural dimensions. The preparatory- and pursuit-related neural dimensions are neither completely orthogonal nor completely overlapping. Thus, there is a reorganization of the population activity in FEF_SEM between preparation and pursuit, but the reorganization is not complete enough to push preparatory activity into a movement-null dimension. In our system, we are able to understand in terms of the responses of its constituent neurons why the population reorganizes partially rather than fully, and we have a functional explanation for why the system might operate in this regime. We identified two subpopulations of cells in our data and one of them (‘Subpopulation 2’) seems to be responsible for the partial reorganization of neural dimensions between preparation and pursuit (Figure 4D). Because the reorganization is not sufficient to prevent movement during preparation, we propose that Subpopulation 1 creates a highly-directional enhancement of visual-motor gain while Subpopulation 2 is responsible for a directionally non-specific component of preparatory visual-motor gain enhancement, in effect providing a rigid and fairly uniform prior for the direction of target motion. A potential benefit of the regime created by the two Subpopulations is that the system will be fully ready for an expected direction of target motion and never be caught entirely flat-footed if target motion occurs in a completely unexpected direction.

Previous work suggests that gating mechanisms play minor roles for pursuit. The discharge of the neurons involved in the switch for saccades is incompatible with the idea that pursuit might use the same switch circuit (Missal and Keller, 2002). The firing of omnipause neurons pauses completely during the execution of saccades of any direction and amplitude but exhibits directionally-tuned decrements of firing rate that are graded according to eye speed during smooth pursuit eye movement. While one could argue that the use of an absolute versus a dynamic gate to prevent activity from causing movement is a matter of semantics, we are suggesting something fundamentally different in this paper: movement is not prevented during modulation of FEF_SEM activity by the action of an inhibitory gate, rather FEF_SEM output is controlling the gate. Indeed, it is possible that FEF_SEM output interacts with omnipause neurons. Omnipause neurons are anatomically connected to at least the saccadic frontal eye fields and their responses during pursuit align closely with properties of visual-motor gain (Langer and Kaneko, 1990; Missal and Keller, 2002; Schwartz and Lisberger, 1994).

Our findings are not due to the trivial conclusion of a lack of importance of FEF_SEM in the generation of pursuit eye movement. Lesioning and inactivating FEF_SEM results in large decrements of pursuit speed and an increase in the number of catch-up saccades (Keating, 1991; Shi et al., 1998). The output of FEF_SEM (and possibly partner areas such as the Supplementary Eye Fields Missal and Heinen, 2001) is necessary to allow visual motion signals to drive the oculomotor machinery (Tanaka and Lisberger, 2001). Furthermore, there are trial-by-trial correlations between FEF_SEM activity during pursuit and eye velocity (Schoppik et al., 2008). The fact that microstimulation of FEF_SEM causes smooth eye movement seems to contradict our conclusion that the output from FEF_SEM is modulatory. We suspect that the artificial situation created by micro-stimulation will drive activity in brainstem oculomotor neurons simply because FEF_SEM is anatomically connected to the motor system. Beyond general anatomical connectivity, micro-stimulation-generated smooth eye movements may say little about how FEF_SEM output is actually used downstream.

We suggest that the brain uses multiple mechanisms to prevent preparatory activity from causing unwanted and premature movement, and that the primary mechanism used by any particular movement system will depend on the organization of that particular system. Further, a system could use multiple mechanisms at the same time, perhaps emphasizing one or another.

The short latency and sensory-driven features of pursuit may mitigate in favor of a system where preparatory activity is used to modulate sensory-motor transmission. In pursuit, movement initiation is driven directly, with relatively short latencies, by sensory inputs from extrastriate visual area MT (Komatsu and Wurtz, 1989; Newsome et al., 1988). Outputs from MT and FEF_SEM converge downstream in the pursuit circuit to control the activity of motoneurons (Lisberger, 2010; Mustari et al., 2009). Importantly, preparatory modulation of visual-motor gain is a genuine preparatory signal. FEF_SEM preparatory responses encode the expected speed and direction of the upcoming target motion and FEF_SEM seems to set the state of the pursuit system during preparation in a way that sculpts the subsequent visual motion signals (Darlington et al., 2018).

In the arm movement system and the motor cortex, it may be difficult to prevent activity in the motor cortex from affecting the firing of motoneurons, and so it is ideal to nullify the preparatory activity by contriving for it to reside in a movement-null dimension. Here, the link from sensory input to motor output is less direct than for pursuit and the output from the primary motor cortex and the premotor cortex affect the firing of motoneurons directly, over monosynaptic and disynaptic pathways (Dum and Strick, 1991; Griffin et al., 2015; Kuypers and Brinkman, 1970). That said, it is possible that gain control mechanisms also operate in arm movements, both to implement the essential function of adjusting to different contexts, and to allow preparatory activity without movement. Just as we think that FEF_SEM is responsible for the preparatory aspects of pursuit and area MT is responsible for the movement drive, recent work in the mouse motor system revealed subpopulations of motor cortex neurons with distinct roles in preparation and movement (Economo et al., 2018). Furthermore, last-order interneurons in the spinal cord show arm-movement-related preparatory activity that could reflect motor cortical modulation of the gain of spinal reflex circuits (Prut and Fetz, 1999).

In saccadic eye movements, the need for high degrees of accuracy and precision in a totally open-loop movement probably requires the preparatory activity that exists at least in the frontal eye fields, the superior colliculus, and the parietal cortex, but also requires an absolute brake on movement until all the brain’s ducks are in a row (Glimcher and Sparks, 1992; Hanes and Schall, 1996; Mazzoni et al., 1996). Here, the switching circuit in the brainstem prevents any movement until the activity that guides the direction and amplitude of the impending saccade is fully prepared to make a movement that is at the same time really quick and accurate (Robinson, 1973). For example, a monkey executes a 20 deg saccade in 40 ms with accuracy of better than 5% (Fuchs, 1967). By contrast, pursuit eye movements are much slower and less accurate, and arm movements rely partly on feedback to achieve maximum accuracy.

Preparatory activity is ubiquitous, but the different features of different motor systems may require different mechanisms to prevent movement during preparation. A switching circuit could exist for movements other than saccades, perhaps in a softer, more compliant form. Gain control, may be an ideal way to establish context sensitivity and adjust the responsiveness to descending signals in systems with short latencies and graded responses. The dynamical system regime may be specialized for neural systems where the neurons involved in preparation make synaptic connections directly onto final motor pathways.

As described in Darlington et al. (2018), two male rhesus monkeys aged 8–10 years and weighing between 12 and 14 kg were used in the neurophysiology experiments (Darlington et al., 2018). The same two monkeys were used for the new behavioral experiments presented here. The monkeys underwent surgeries to implant head-restraining hardware onto the skull and a scleral search coil to track eye movements. Following recovery, monkeys were trained to fixate and smoothly track moving visual targets. During experiments, analog signals produced by the scleral search system recorded vertical and horizontal eye position. Eye velocity-related signals were obtained by passing the position-related signals through an analog circuit that differentiated signals from DC to 25 Hz and rejected signals at higher frequencies (−20 dB/decade). Eye position and velocity signals were sampled at 1 kHz and stored for offline data analysis. After training, monkeys underwent another surgery in which a craniotomy was performed in an area centered over the smooth eye movement region of the frontal eye fields (FEF_SEM). A sealable, titanium chamber was fixed to the skull over the craniotomy. All procedures received prior approval by Duke’s Institutional Animal Care and Use Committee and complied with the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals.

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

We thank Stefanie Tokiyama, Bonnie Bowell, Scott Ruffner, and Steven Happel for technical assistance, and the other members of our laboratory for helpful comments on the manuscript. We thank J Patrick Mayo for providing the figure panel found in our response to the reviews. Research supported by R01-EY027373 (SGL) and F30-EY027684 (TRD).

Animal experimentation: All procedures received prior approval by Duke's Institutional Animal Care and Use Committee (protocol A085-18-04) and were in compliance with the National Institutes of Health's Guide for the Care and Use of Laboratory Animals.

© 2020, Darlington and Lisberger

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.