Responses of neurons in macaque MT to unikinetic plaids

Abstract

Response properties of MT neurons are often studied with "bikinetic" plaid stimuli, which consist of two superimposed sine wave gratings moving in different directions. Oculomotor studies using "unikinetic plaids" in which only one of the two superimposed gratings moves suggest that the eyes first move reflexively in the direction of the moving grating and only later converge on the perceived direction of the moving pattern. MT has been implicated as the source of visual signals that drives these responses. We wanted to know whether stationary gratings, which have little effect on MT cells when presented alone, would influence MT responses when paired with a moving grating. We recorded extracellularly from neurons in area MT and measured responses to stationary and moving gratings, and to their sums: bikinetic and unikinetic plaids. As expected, stationary gratings presented alone had a very modest influence on the activity of MT neurons. Responses to moving gratings and bikinetic plaids were similar to those previously reported and revealed cells selective for the motion of plaid patterns and of their components (pattern and component cells). When these neurons were probed with unikinetic plaids, pattern cells shifted their direction preferences in a way that revealed the influence of the static grating. Component cell preferences shifted little or not at all. These results support the notion that pattern-selective neurons in area MT integrate component motions that differ widely in speed, and that they do so in a way that is consistent with an intersection-of-constraints model.NEW & NOTEWORTHY Human perceptual and eye movement responses to moving gratings are influenced by adding a second, static grating to create a "unikinetic" plaid. Cells in MT do not respond to static gratings, but those gratings still influence the direction selectivity of some MT cells. The cells influenced by static gratings are those tuned for the motion of global patterns, but not those tuned only for the individual components of moving targets.

Keywords: MT; direction selectivity; electrophysiology; motion integration; motion perception.

Fig. 1.

Intersection of constraints (IOC) in velocity space, illustrated for bikinetic and unikinetic plaid stimuli. The perceived motion of visual patterns can be understood within an IOC framework. An individual grating is perceived as moving orthogonal to its orientation, depicted by the black arrows, but this stimulus is also consistent with any number of faster-moving gratings that share the same orthogonal component, indicated by gray arrows. The end points of these arrows, taken together, form a constraint line in velocity space (dashed lines). Connecting the origin of this velocity space with the point where the constraint lines intersect yields the vector (red arrow) that corresponds to the true motion of the stimulus, and to the percept of an unbiased observer who sees the two gratings superimposed. In the case of bikinetic plaids (A), two moving gratings of equal speed are superimposed, so the constraint lines intersect on the horizontal axis, giving the percept of a stimulus moving to the right. In the unikinetic case (B), one moving and one static grating are superimposed. The constraint line imposed by the static grating goes through the origin, shifting the intersection point of the constraint lines 30° off the horizontal, toward the veridical motion vector of the moving grating. The lengths of the red arrows indicate that the unikinetic plaid appears to move more slowly than the bikinetic one.

Fig. 2.

Example responses of pattern and component cells to gratings and plaids. A: curves in the center of each stimulus family show the predicted responses of idealized pattern-selective (red) and component-selective (blue) neurons tuned for rightward motion when tested with the stimuli depicted in the ring. These correspond to the conditions used in our experiment. All conditions were tested in 12 directions, although for economy the full set is only shown for the static case. Each column shows responses to a different set of targets. Stationary gratings: MT neurons are generally not strongly responsive or selective for static patterns, so we expect only weak responses, not differing between pattern and component cells and indicated in black. Moving gratings: we expect both pattern and component cells to exhibit a robust unimodal response, represented by the black von Mises function in the center. The motion of each grating is indicated by black arrows. Bikinetic plaids: as in Fig. 1A, these consist of two moving component gratings (the motion of which is indicated by blue arrows) yielding one pattern motion (red arrows). We expect pattern and component cells to exhibit bimodal (component cell) and unimodal (pattern cell) tuning, represented by blue and red von Mises functions. Left- and right-handed unikinetic plaids: as in Fig. 1B, these consist of a moving grating and a stationary one; the left- and right-handed cases differ in which grating is moving. The color scheme is the same as for the bikinetic plaid. We expect both pattern and component cells to exhibit unimodal tuning. The component cells should respond to the moving grating essentially as if the static grating were absent. We therefore expect a difference in preferred direction of 60° between the two kinds of unikinetic plaid. Pattern cells’ tuning should be aligned with the pattern motion, and the difference in preferred directions should be roughly 0°. B–E: responses of 4 representative example cells. Color scheme is same as in A, but pattern and component predictions are depicted by dashed lines, whereas measured responses are solid lines. All cells responded weakly to static gratings and showed classical response characteristics to moving gratings and bikinetic plaids. The component cell (B) responded in a fashion consistent with the component prediction (direction tuning rotation was 52°). The pattern cells (D and E) obeyed the pattern prediction, although not perfectly (direction tuning rotation values of 7° and 16°, respectively). The mixed cell (C) showed classical behavior to moving gratings and bikinetic plaids and would be classified as a pattern cell when stimulated with bikinetic plaids, but when probed with unikinetic plaids, it showed component cell-like behavior (direction tuning rotation of 38°).

Fig. 3.

Pattern/component classification for bikinetic and unikinetic plaids. A: bikinetic plaids. The ordinate represents the partial correlation of the neural response with the pattern prediction (Z_p), and the abscissa represents the partial correlation of neural response with the component prediction (Z_c). Black lines indicate the borders of significance that allow a statistical classification of cell behavior. Pattern cells (falling in upper region) are red, and component cells (falling in lower region) are blue; unclassed cells (falling between these regions) are black. Open circles indicate that the cell was recorded in the acute preparation, and closed circles indicate that the cell was recorded under awake conditions. Large black circles indicate the example cells from Fig. 2. B: unikinetic plaids (averaged over both left- and right-handed unikinetic conditions). The cells are colored based on their classification determined with bikinetic plaids.

Fig. 4.

Pattern index predicts the rotation of direction preference induced by static stimuli. The abscissa gives the pattern index (Z_p − Z_c) for bikinetic plaids, and the ordinate gives the direction preference difference between responses to left- and right-handed unikinetic plaids. Data from the alert animal are depicted in orange, and data from the anesthetized animals are represented in green; error bars represent the bootstrapped standard deviations of the estimates for pattern index and rotation. Solid lines represent the lines of best fit (Press et al. 1992) for the two groups of cells. Arrows show the direction tuning rotation expected from idealized pattern and component cells (Fig. 2A). Black circles indicate the example cells from Fig. 2. Shaded background colors indicate the ranges of pattern and component cells as displayed in Fig. 3.

Fig. 5.

Time course of responses and tuning preferences for asynchronously presented compound stimuli. A: experimental design. At the start of all trials, we introduced a static grating (gray arrow). For trials with compound stimuli, we added a moving grating with an orientation 60° different from the static grating either synchronously (red) or with a delay of 50 or 100 ms (blue or cyan, respectively). B and C: mean firing rates evoked by optimal single gratings and plaids containing those gratings for 121 cells from the alert animal (B) and for 84 cells from the anesthetized animal (C) for the 4 stimulus conditions schematized in A, computed at 1-ms intervals within a 5-ms sliding window. D and E: mean direction tuning to moving stimuli for the neuronal populations (alert, D; anesthetized, E). We rotated the tuning data for all neurons so that the preferred direction for single gratings was aligned at horizontal (indicated as 30°) and estimated the population preferred direction within the same 5-ms sliding window. The preferred direction is plotted only for times past the time at which the estimate for each stimulus had stabilized.