TY - JOUR
AU - Gu, Yong
AB - Abstract Accurate heading perception relies on visual information integrated across a wide field, that is, optic flow. Numerous computational studies have speculated how local visual information might be pooled by the brain to compute heading, but these hypotheses lack direct neurophysiological support. In the current study, we instructed human and monkey subjects to judge heading directions based on global optic flow. We showed that a local perturbation cue applied within only a small part of the visual field could bias the subjects’ heading judgments, and shift the neuronal tuning in the macaque middle temporal (MT) area at the same time. Electrical microstimulation in MT significantly biased the animals’ heading judgments predictable from the tuning of the stimulated neurons. Masking the visual stimuli within these neurons’ receptive fields could not remove the stimulation effect, indicating a sufficient role of the MT signals pooled by downstream neurons for global heading estimation. Interestingly, this pooling is not homogeneous because stimulating neurons with excitatory surrounds produced relatively larger effects than stimulating neurons with inhibitory surrounds. Thus our data not only provide direct causal evidence, but also new insights into the neural mechanisms of pooling local motion information for global heading estimation. heading perception, macaque, microstimulation, MT, optic flow Introduction When navigating in the world, the surrounding objects generate a radial pattern of flow on our retina, the so-called “optic flow” (Gibson 1950). The focus of expanding flow pattern (FOE), provides critical information to the brain for computing the heading direction of a person moving through the environment. Both human and nonhuman primates can reliably discriminate heading direction as finely as a few degrees based on optic flow (Warren et al. 1988; Warren and Saunders 1995; Royden and Hildreth 1996; Britten and van Wezel 1998; Britten 2008; Gu et al. 2008). On the other hand, in nature, this process is often complicated by the interference of the signals that are independent on self-motion, for example, those moving objects that may generate local motion components superimposed onto the observer-generated optic flow. If the brain pools information from every part of the visual field, then the final heading estimation would likely be biased by these local motion perturbations. This indeed has been shown as in a number of psychophysical experiments (Duffy and Wurtz 1993; Warren and Saunders 1995; Royden and Hildreth 1996; Dokka et al. 2015; Layton and Fajen 2016). In regard to neural mechanisms that may underlie heading estimation with and without these perturbation signals, modeling studies have proposed networks involving visual motion neurons like those in MT, and in its downstream area—the dorsal portion of the medial superior temporal area (MSTd) (Warren and Saunders 1995; Royden 2002; Layton et al. 2012). Specifically, the relatively small receptive fields of area MT neurons (compared with MSTd) are capable of encoding motion information in each local visual field of each eye. Downstream areas, for example, MSTd neurons, which possess much larger receptive fields, then pool the signals, and serve as templates for processing complex optic flow patterns with different FOE locations. These models have offered plausible neural mechanisms underlying heading computation, and have generated predictions that may match behavioral observations such as the biased heading perception effect under presence of local perturbation cues. However, there are 2 important issues left unsolved: 1) Direct neurophysiological evidence supporting this hypothesis is basically lacking. In particular, the contributions of the motion direction signals in area MT to final heading computation under the context of optic flow have never been experimentally proved; 2) It has been reported that MT contains different types of neurons with surround inhibition or facilitation effect, and researchers have speculated that they might be related to local object-ground segregation and wide field-motion processing, respectively (Tanaka et al. 1986; Born and Tootell 1991, 1992; Born et al. 2000; Born 2000). How these different types of neurons are exactly involved in a fine heading discrimination task is unknown. To address these issues, we instructed both human subjects and macaques to perform a fine heading direction discrimination task based on global optic flow (Gu et al. 2008; Fetsch et al. 2011; Chen et al. 2013). In some of the trials, a part of the local field was perturbed with extra motion vectors either leftward or rightward that were independent on the original flow due to self-motion. While measuring the behavior, we recorded neuronal activity from MT neurons in 2 macaques by using tungsten electrodes in conjunction with electrical microstimulation (Salzman et al. 1990; Salzman et al. 1992; Murasugi et al. 1993). We showed that similar to the local visual perturbation effects on the behavioral performance, stimulating MT neurons with receptive fields that occupied only a small portion (~5° × 5°) of the wide visual field (~90° × 90°) could also significantly influence the subjects’ heading judgments based on global optic flow. Materials and Methods Psychophysics A total of 6 human subjects and 2 rhesus monkeys (Macaca mulatta) participated in the study. Human subjects except the author were naive to the current study. During the task, human subjects were seated in a chair with their heads fixed on a cushioned holder and were instructed to keep fixation. Monkeys were seated in a custom-built primate chair with their heads restrained by a chronically implanted circular molded lightweight plastic ring (5 cm in diameter). The ring was anchored to the skull using titanium inverted T-bolts and dental acrylic (Gu et al. 2006) . Both monkeys were implanted with scleral coils for measuring eye movements in a magnetic field (Riverbend Instruments). Behavioral training for monkeys was accomplished using standard operant conditioning procedures (Gu et al. 2008,2012). All procedures were approved by the Animal Care Committee of Shanghai Institutes for Biological Science, Chinese Academy of Science (Shanghai, China). System and Setup Subjects viewed visual stimuli on a large LCD monitor (Samsung ED55C for human subjects, 60 Hz, HP LD4201 for monkey subjects, 60 Hz). The pixel-on/off delay was measured and defined as the interval between baseline and 95% of evoked signals reaching a plateau from a photodiode attached on the screen. According to this measurement, the pixel-on delay was ~41 ms for HP LD4201, and ~30 ms for Samsung ED 55 C. The pixel-off delay was 24 ms for HP LD4201, and ~27 ms for Samsung ED 55 C. The visual display for the monkeys was viewed from a distance of 30 cm, subtending a visual angle of ~90° × 90°. For human subjects, the viewing distance was 57 cm and the viewing angle was a bit smaller (~60° × 60°). The screen had a resolution of 1980 ×1080 pixels with a refresh rate of 60 Hz. Visual stimuli were generated by an OpenGL accelerator board (nVidia Quadro 2000) and were plotted with subpixel accuracy using hardware antialiasing. Behavioral tasks and data acquisitions were controlled by Tempo software (Reflective Computing, Olympia, WA). Visual Stimuli Optic flow simulating self-translation through a 3D cloud of stars was distributed uniformly within a virtual space of 100 cm wide, 100 cm high and 40 cm deep (corresponding to roughly 90° × 90° × 40°). The density of the stars was 0.01 /cm3, and each of the stars was a 0.2 cm × 0.2 cm triangle. The stimuli contained a variety of depth cues, including motion parallax and size information. Motion coherence was manipulated by randomizing the 3D location of a percentage of stars on each frame while the remaining stars moved coherently (Gu et al. 2012). In particular, the coherence was about 15~20% and under these conditions, the phychophysical thresholds of the subjects were roughly a few degrees (1~3°). All these parameters were similar as used in the previous experiments (Gu et al. 2008,2012), allowing qualitative comparison of the results in different studies. Heading Discrimination Task Subjects were trained to perform a fine heading discrimination task (Gu et al. 2008). They were instructed to fixate a central target at the beginning of each trial. After a delay of 250 ms’ the 3D-cloud of stars was presented, eliciting a percept of forward motion with a small degree of leftward or rightward component (0°, ±0.51°, ±1.28°, ±3.2°, and ±8°). Each simulated heading condition was repeated for at least 10 times, leading to ≥90 trials in one experimental block. To make better comparison with the previous results reported in area MSTd (Gu et al. 2008), the motion trajectory followed a Gaussian velocity profile to simulate real motion (traveling distance: 20 cm; peak velocity: 30 cm/s). Each motion presentation lasted 1 second. Subsequently, subjects reported their perceived heading (“left” vs. “right”). Human observers pressed a key on a keyboard, and a beep at the end of a trial signaled whether the response was correct. Monkeys made a saccadic eye movement to one of the 2 targets and received a drop of water, when the response was correct. Note that a correct choice was assigned when it was consistent with the background optic flow, regardless of any local perturbation or microstimulation signals. In all of the experiments on monkeys, during the fixation period (1 s), the trial was stopped and discarded if the animals’ conjugate eye position deviated from a 2° × 2° electronic window centered at the fixation point. The fixation point was always fixed at the center of the screen, thus did not require the subjects to make any smooth eye movements in the presence of the optic flow stimuli (Britten and Van Wezel 2002). At the end of the trial, the choice targets appeared 10° eccentric to the central screen after the fixation point disappeared. The subjects were required to make a choice within 500 ms. Local Perturbation Cues After the subjects reached stable performance in the heading discrimination task, 2 local perturbation cues (one leftward, and the other rightward) were introduced into the heading stimuli, that is, optic flow. These perturbed trials occupied two-thirds of the total trials and were randomly interleaved with the non-perturbed trials, leading to a full block containing at least 270 trials (90 × 3). The perturbed area was either centered on th screen, or was located in peripheral areas (see Supplementary materials). The size of the perturbation area was randomly chosen from windows extending 2° × 2°, 4° × 4°, 6° × 6°, 8° × 8°, 10° × 10°, 12° × 12°, and 15° × 15° (no 12° × 12° for human subjects), with identical depth as the background flow (40°). Each size parameter was repeated for at least 3 times in human subjects (6 × 3 = 18 blocks), and 5 times in monkeys (7 × 5 = 35 blocks). The trajectories of all the dots located within the perturbed area were biased with a motion vector towards leftward or rightward. The length of the ‘added’ vector was set to be 1/10 of the original vector, leading to a deviation of 6° away from the self-motion trajectory within the perturbed area (α in Fig. 1A). In the control experiments, we also increased the salience of the perturbed area from the background flow by changing its color (see Supplementary materials). Figure 1. View large Download slide Behavioral paradigms. (A) Cartoon illustrates the motion vector when the observer moves (black arrow) or when the object moves (red arrow). The resultant vector is depicted by the dashed red arrow, which represents the vector summation of the 2 motion components. In the experiment, the ratio between the red and black arrow is 1/10, leading to the deviated angle (α) about 6°. (B) Optic flow pattern used to simulate self-motion, i.e., heading of the observer in the forward direction (left panel), and self-motion with a central patch of leftward (middle panel) or rightward deviated flow (right panel). The color in the plots is for demonstration only. (C) Flow chart of events in psychophysical and electrical microstimulation experiments. Monkeys reported the perceived heading (“left” vs. “right”) by making a saccadic eye movement to one of 2 lateral targets (left and right, respectively), while human observers pressed a key on the keyboard. In one half of the trials electrical microstimulation was randomly interleaved with non-stimulation control trials (Gu et al. 2012). (D, E) Psychometric functions plotting proportion of rightward responses (corresponding to “rightward” heading) as a function of heading direction for one human and one monkey subject, respectively. Black: self-motion; red: self-motion with leftward (L) component added; green: self-motion with rightward (R) component added. The size of the central window within which a directional bias of object motion was added to the optic flow pattern was 6° × 6°.The arrows underneath the x-axis represent the optic flow patterns associated with each heading directions. Positive values in the abscissa represent rightward heading and negative values represent leftward heading. (F, G) Population results of the induced PSE shift as a function of the perturbed area size for 6 human and 2 monkey subjects (black curves), and from the model prediction (grey curves, see detail in Supplementary Fig. 8). The positive value of the y-axis refers to the PSE shift in the expected direction from the perturbation vector. Error bars: standard error of the mean, SEM. Figure 1. View large Download slide Behavioral paradigms. (A) Cartoon illustrates the motion vector when the observer moves (black arrow) or when the object moves (red arrow). The resultant vector is depicted by the dashed red arrow, which represents the vector summation of the 2 motion components. In the experiment, the ratio between the red and black arrow is 1/10, leading to the deviated angle (α) about 6°. (B) Optic flow pattern used to simulate self-motion, i.e., heading of the observer in the forward direction (left panel), and self-motion with a central patch of leftward (middle panel) or rightward deviated flow (right panel). The color in the plots is for demonstration only. (C) Flow chart of events in psychophysical and electrical microstimulation experiments. Monkeys reported the perceived heading (“left” vs. “right”) by making a saccadic eye movement to one of 2 lateral targets (left and right, respectively), while human observers pressed a key on the keyboard. In one half of the trials electrical microstimulation was randomly interleaved with non-stimulation control trials (Gu et al. 2012). (D, E) Psychometric functions plotting proportion of rightward responses (corresponding to “rightward” heading) as a function of heading direction for one human and one monkey subject, respectively. Black: self-motion; red: self-motion with leftward (L) component added; green: self-motion with rightward (R) component added. The size of the central window within which a directional bias of object motion was added to the optic flow pattern was 6° × 6°.The arrows underneath the x-axis represent the optic flow patterns associated with each heading directions. Positive values in the abscissa represent rightward heading and negative values represent leftward heading. (F, G) Population results of the induced PSE shift as a function of the perturbed area size for 6 human and 2 monkey subjects (black curves), and from the model prediction (grey curves, see detail in Supplementary Fig. 8). The positive value of the y-axis refers to the PSE shift in the expected direction from the perturbation vector. Error bars: standard error of the mean, SEM. Electrophysiological Recordings Extracellular action potentials were recorded by using single Tungsten microelectrodes (FHC; impedance, ~500 k). During each experiment, the microelectrode was inserted into the cortex through a transdural guide tube, using a hydraulic Microdrive (FHC). The signal from the electrode was amplified (Bak Electronics), filtered (400 Hz to 5 kHz), digitized (25 kHz), and stored to disk. Single units were isolated off-line using Spike2 (Cambridge Electronic) via template matching as well as PCA methods. Anatomical Localization of Recordings Area MT was identified by magnetic resonance imaging scans, stereotaxic coordinates (~15 mm lateral and 3–6 mm posterior to the interaural axis), and physiological response properties. MT neurons contained small contralateral receptive fields (average: 5.5°) and strong responses to visual stimuli (Dubner and Zeki 1971; Van Essen et al. 1981; Desimone and Ungerleider 1986; Komatsu and Wurtz 1988). Receptive fields, including the preferred direction and speed of the isolated neurons were first hand-mapped with a patch of drifting random dots, and then were more quantitatively defined by scanning the visual display with a small bar (1~2° × 1~2°) moving in the preferred direction at the preferred speed. The penetrations were conducted from the top such that in the majority of penetrations, we first encountered MSTd before MT. This pattern further confirmed that we were in the correct areas (Gu et al. 2006). We recorded from any site within these areas that was spontaneously active or responded to a small pattern of flickering or moving random dots. Thus, our sample of neurons was unbiased with respect to direction preference. Microstimulation Procedures We first recorded multiunit (MU) activity by single tungsten electrodes (FHC; impedance, ~500 k). Along each penetration, 2 types of tuning curves of the isolated units were mapped at multiple depths parted by 100μm. First, tuning curves under wide field stimuli were measured in the horizontal plane (10 directions relative to straight ahead: 0°, ±22.5°, ±45°, ±90°, ±135°, and 180°) with optic flow stimuli occupying the whole screen. Second, tuning curves under random dots stimuli restricted inside the isolated unit's receptive fields were measured in the vertical, that is, frontoparallel plane. Each stimulus condition was repeated for more than 3–5 trials. A recording site was chosen for microstimulation if its tuning under optic flow was similar to that of neighboring sites (±100 μm, i.e., above or below the recording location). A low-amplitude electrical current was delivered (Alpha Omega SnR) through the tip of the tungsten microelectrode (200 Hz, biphasic, cathodal leading, pulse width = 200 μs for each phase duration, inter pulse interval = 100 μs). The current was modulated by a Gaussian envelope with a peak amplitude of 20μA (see Fig. 1C, “electrical current”) because: 1) the visual responses mainly follow the velocity profile and 2) to be consistent with previous research regarding both the envelope and the peak of the current amplitude (Gu et al. 2012), thus allowing us to make fair comparisons of the results across studies. In some conditions, the peak amplitude of the current was set to 10, 40, 60, and 200 μA. Trials with microstimulation were randomly interleaved with the nonstimulated trials. The animals were always rewarded according to the actual heading stimuli. Data Analysis Data analyses were performed using MATLAB (Mathworks). To quantify behavioral performance, the proportion of ‘rightward’ choices was plotted as a function of heading directions, and the resulting psychometric functions were fitted with a cumulative Gaussian function. The psychophysical threshold, taken as the standard deviation of the Gaussian fit, corresponded to 84% correct performance. The point of subjective equality (PSE) was taken as the mean of the Gaussian fit, corresponding to the heading angle that yielded an equal percentage of rightward or leftward responses (Gu et al. 2008, 2012) . Each stimulus condition was repeated for at least 10 trials. Tuning Strength To quantify the overall strength of the direction tuning of MT neurons under optic flow, we computed the direction discrimination index (DDI) as defined previously (Prince et al. 2002; Chen et al. 2011):   DDI=(Rmax−Rmin)Rmax−Rmin+2×SSEN−M, (1) where Rmax and Rmin are the mean firing rates of the neuron along the optic flow-simulated heading directions that elicited maximal and minimal responses, respectively. SSE is the sum-squared error around the mean responses, N is the total number of observations (trials), and M is the number of stimulus directions. This index quantifies the amount of response modulation (due to changes in stimulus direction) relative to the noise level. Neurons with large response modulations relative to the noise will have DDI values close to 1, whereas neurons with weak response modulations relative to noise level will have DDI values close to 0. Surround Index To measure the center-surround interaction of MT neurons, we computed a surround index (SI) (Born et al. 2000; Born 2000):   SI=RWF−RSFRWF+RSF, (2) where RWF represented responses to the wide field stimulus and RSF were responses to the stimulus restricted within the receptive field. SI approaching −1 indicated strong antagonistic surround, while values approaching one indicated strong excitatory surrounds. A value of 0 indicated neither inhibitory nor excitatory surround effects. Results Heading Performance without and with Local Perturbation A total of 6 human subjects and 2 monkeys were tested with a fine heading discrimination task (see more detail in Methods) in which global optic flow (subtending ~90° × 90°) elicited perception of a linear translation of the observer's approach towards a visual display (Fig. 1A; left patch in Fig. 1B). To assess the influence of local motion perturbation on heading estimation, in some of the trials an extra local motion component, leftward or rightward, was introduced to the background dots within a certain restricted area, leading to locally distorted flow (Fig. 1A; middle and right patches in Fig. 1B). All trials in the 3 experimental conditions were randomly interleaved in one experimental block (Fig. 1C). In all conditions, human subjects were instructed to judge their heading directions from the wide-field flow by ignoring the local perturbation signals. For monkey subjects, they were rewarded according to the heading directions defined by optic flow outside the perturbed area. Specifically, subjects chose a “left” target if the global optic flow-simulated heading direction was leftward, and vice versa for the rightward heading directions (see Method). Figure 1D and E shows heading performance as quantified by psychometric functions from one example block in human (D) and monkey subjects (E), respectively. Consistent with earlier findings, the subjects could reliably discriminate leftward from rightward heading directions relative to straight-ahead based on optic flow (Gu et al. 2008, 2010, 2012). However, in trials with a central window of local motion (6° × 6° in this example) superimposed onto the global optic flow pattern, there was a significant bias in perceived heading direction towards the opposite side of the introduced motion vector as quantified by the shift of the PSE in both human (|ΔPSE, L-perturb vs. no-perturb| = 0.99°, P = 0.034; |ΔPSE, R-perturb vs. no-perturb| = 1.44°, P = 1.7E-4, probit regression) and monkey subjects (|ΔPSE, L-perturb vs. no-perturb| = 1.26°, P = 4.0E-4;|ΔPSE, R-perturb vs. no-perturb| = 1.65°, P = 1.3E-7, probit regression). Specifically, subjects made more rightward choices, when an extra leftward local motion component (red vs. black symbols in Fig. 1D and E) was introduced and vice versa for an added rightward local motion component (green vs. black symbols in Fig. 1D and E). Intuitively, the impact of this local motion perturbation on the global heading estimation may depend on its area size relative to the whole visual field. This is indeed the case as shown in Figure 1F and G. At the population level, the shifted heading PSE induced by the perturbation cue grew monotonically as a function of its size for both human (human 1: R = 0.99, P = 2.3E-4; human 2: R = 0.96, P = 0.0027, human 3: R = 0.91, P = 0.012; human 4: R = 0.94, P = 0.0047, human 5: R = 0.86, P = 0.027; human 6: R = 0.96, P = 0.0027, Pearson correlation coefficient) and monkey subjects (monkey R: R = 0.97, P = 2.4E-4; monkey Y: R = 0.89, P = 0.016, Pearson correlation coefficient). Notably, perturbation of motion within an area as small as 4° × 4° visual angle that occupied <1% of the background field could already affect the subject's heading perception based on global optic flow (P < 0.05, two tailed t-test). Furthermore, we examined other factors including eccentricity, saliency, and training period of the perturbation cues, we found that the biased effect on the perceived heading was robust in all these control conditions (see Supplementary Fig. 1). Hence, these behavioral results suggest that the introduced local perturbation signals are more likely to be treated by the brain as part of the background flow induced from self-motion. Such data are consistent with a spatial pooling mechanism, by which the visual channel pools all local motion vectors to compute FOE for heading estimation (Fig. 1F and G), as observed in the previous psychophysical studies (Warren and Saunders 1995; Dokka et al. 2015; Layton and Fajen 2016). In the following, we explore how this process is mediated by neurons in the brain, in particular, the MT cells that are sensitive to visual motion cues within a restricted area in the visual field. MT's Responses to Optic Flow and Local Perturbation Cues We recorded 305 MT units in the above 2 monkeys. In the following sections, we first examined neural properties from 64 well isolated single-unit (SU, Fig. 2), and then finally examined the rest 241 multiunits to see whether they show similar trends (MU, Supplementary Fig. 2). Figure 2. View largeDownload slide Responses of MT single units to optic flow. (A) Distribution of recorded MT single neurons’ receptive fields (RFs). Red color represents units recorded in the right hemisphere whereas the blue represents units recorded in the left hemisphere. (B) An example MT unit's responses to optic flow with FOE varied in the whole horizontal plane (arrows underneath the abscissa indicate the flow motion directions). (C) Population results for 64 neurons of the relationship between the direction discrimination index (DDI) and the surround index (SI) for small patch (red symbols) and whole-field stimulation (black symbols). Neurons with excitatory surrounds tend to be more sharply tuned (larger DDI) to whole-field stimuli. The red and black curves represent running mean values with standard error bars. (D) Population tuning curves (n = 21) of MT neuronal responses in the heading discrimination task without perturbation (open symbols) and with perturbation signals in the neurons’ preferred direction (black solid symbols). Error bars: SEM. Figure 2. View largeDownload slide Responses of MT single units to optic flow. (A) Distribution of recorded MT single neurons’ receptive fields (RFs). Red color represents units recorded in the right hemisphere whereas the blue represents units recorded in the left hemisphere. (B) An example MT unit's responses to optic flow with FOE varied in the whole horizontal plane (arrows underneath the abscissa indicate the flow motion directions). (C) Population results for 64 neurons of the relationship between the direction discrimination index (DDI) and the surround index (SI) for small patch (red symbols) and whole-field stimulation (black symbols). Neurons with excitatory surrounds tend to be more sharply tuned (larger DDI) to whole-field stimuli. The red and black curves represent running mean values with standard error bars. (D) Population tuning curves (n = 21) of MT neuronal responses in the heading discrimination task without perturbation (open symbols) and with perturbation signals in the neurons’ preferred direction (black solid symbols). Error bars: SEM. Consistent with previous findings (Allman and Kaas 1971; Dubner and Zeki 1971; Van Essen et al. 1981), MT neurons usually contained relatively smaller receptive fields compared with those in its downstream area MSTd (Fig. 2A), and their sizes increased with increasing eccentricity (R = 0.84, P = 0, Spearman rank correlation). After identifying MT neurons, we then examined how these neurons with small receptive fields respond under the global optic flow stimuli that are typically experienced during self-motion. This is not immediately clear because MT neurons may have inhibitory or excitatory surrounds, which could suppress or enhance the response if the stimulus exceeds their receptive field (Tanaka et al. 1986; Born and Tootell 1992; Raiguel et al. 1995; Berezovskii and Born 2000; Born 2000). Figure 2B shows an exemplary neuron preferring optic flow moving forward. The responses are reduced somehow to the whole stimulus field compared with a much smaller stimulus within its receptive field (~5° × 5°). To quantify this effect, we used a surround inhibition index (SI) with negative values indicating inhibition and positive values indicating excitation (see Methods). As for the neuron shown in Figure 2B, the SI was −0.32, revealing an inhibition effect. Across the population (Fig. 2C), we found that for MT neurons with inhibitory surrounds (e.g., SI < 0), their tuning strength (DDI, see Methods) was slightly reduced for the wide visual field compared with the small patch condition (whole field: 0.61 ± 0.02; small patch: 0.66 ± 0.02, mean ± SEM, P = 0.0015, two tailed paired t-test). In contrast, in neurons with excitatory surrounds, tuning strength to the global optic flow tended to be larger than to the small patch stimuli (whole field: 0.76± 0.02; small patch: 0.65 ± 0.02, mean ± SEM, P = 5.75E-05, two tailed paired t-test). These data demonstrate that in general, MT neurons either with inhibitory or excitatory surrounds are robustly activated during self-motion when subjects experience global optic flow stimuli. Overall, both the response strength and the proportion of neurons in MT that were significantly tuned to optic flow was slightly lower, yet analogous to that in the downstream MSTd neurons that were measured from the same 2 monkeys (MT: 0.66 ± 0.01, 83%; MSTd: 0.71 ± 0.01, 93%, mean ± SEM, Fig. 2C, marginal distributions). These data suggest that the motion direction information in MT is largely preserved when propagated to downstream areas. As described above in the psychophysical results, heading judgments are usually interfered and biased when the optic flow cues are distorted in part of the visual field. The small receptive fields of MT cells (relative to the whole visual field) would make them suitable to detect these perturbation signals during navigation. To test this idea, we introduced a perturbation signal that would result in locally distorted flow with added vectors in the neuron's preferred direction. To guarantee persistent stimulation of MT neurons, this perturbed flow was restricted within the recorded unit's receptive field through the whole stimulation period of time. Such an effect could be clearly seen in the population tuning curves (Fig. 2D): compared with the control (open symbols), the neuronal responses on average increased by ~7.05% (P = 8.45E-04, intercept effect in ANCOVA analysis, black solid symbols). Hence, MT neurons are highly sensitive to motion vectors induced not only from self-motion, but also from local perturbation cues such as those independently moving objects across the visual field during natural navigation. Finally, we reperformed the above analyses on 241 multiunits, and found very similar properties as SU (see Supplementary Fig. 2). This result indicates that there is a highly clustered structure of neuronal responses in MT, allowing us to conduct the microstimulation experiments that are largely based on MU properties in the following sections. Causal Roles of MT Involved in Neural Circuit for Heading Computation To examine whether the MT directional signals were indeed propagated to downstream areas for further heading computation, we introduced a weak current (peak amplitude: 20 μA) into MT to artificially perturb its activity. In each block, only half of the trials were applied with microstimulation and they were interleaved with the other half of trials without microstimulation. A precondition for this technique to work is that the manipulated signals around the electrode tip need to be consistent. Previous work has shown that MT neurons are arranged in a column or cluster (Dubner and Zeki 1971; Maunsell and Newsome 1987; Salzman et al. 1992; Born and Bradley 2005). This is also the case as measured in our own data: neighboring sites along each electrode penetration tended to exhibit similar tuning properties under both wide-field optic flow and local random dots stimuli restricted inside the receptive fields (Supplementary Fig. 3). Figure 3A shows that neurons along electrode penetrations in 3 MU sites when mapped at an interval of 100 μm all preferred rightward moving optic flow (arrows in left panels). Applying electrical stimulation around these sites presumably will add more rightward motion signals of optic flow to the neural circuitry, and thus drive the subjects to report simulated heading direction more frequently to leftward. Our data are consistent with this net excitatory effect: stimulating the middle site significantly biased the animal's heading choice towards more leftward (ΔPSE = 2.3°, P = 1.8E-5, probit regression, Fig. 3B). Moreover, recordings after a full block of microstimulation experiment confirmed that tuning curves of the manipulated neurons remained largely unchanged (Supplementary Fig. 5), suggesting that the applied electrical currents only temporarily influenced neural activities without producing significant harmful effects. Figure 3. View largeDownload slide Effects of microstimulation of MT units (peak amplitude = 20 μA). (A) Examples of tuning curves during one penetration. Each penetration mapped the direction selectivity of 3 MU sites at an interval of 100 µm. These MUs consistently preferred rightward flow (indicated by the arrows in the left panels), that is, simulating leftward heading. (B) One example of microstimulation experiments as derived from the middle sites in (A). Proportion of rightward responses plotted as a function of heading direction. Red: trials using microstimulation; Black: trials without microstimulation. (C, D) Population distribution of induced PSE shift of MT neurons for monkey R and monkey Y, respectively. Positive values represent shifts in the direction predicted from the heading tuning of the stimulated neurons. Filled columns represent significant and open columns insignificant PSE shifts as assessed by probit fit regression. Arrows, Mean PSE shift. The dashed vertical lines mark a zero PSE shift. Figure 3. View largeDownload slide Effects of microstimulation of MT units (peak amplitude = 20 μA). (A) Examples of tuning curves during one penetration. Each penetration mapped the direction selectivity of 3 MU sites at an interval of 100 µm. These MUs consistently preferred rightward flow (indicated by the arrows in the left panels), that is, simulating leftward heading. (B) One example of microstimulation experiments as derived from the middle sites in (A). Proportion of rightward responses plotted as a function of heading direction. Red: trials using microstimulation; Black: trials without microstimulation. (C, D) Population distribution of induced PSE shift of MT neurons for monkey R and monkey Y, respectively. Positive values represent shifts in the direction predicted from the heading tuning of the stimulated neurons. Filled columns represent significant and open columns insignificant PSE shifts as assessed by probit fit regression. Arrows, Mean PSE shift. The dashed vertical lines mark a zero PSE shift. To evaluate effects of population microstimulation, we computed ΔPSE, that is, the induced PSE shift for each experiment, and assigned a positive sign if the shifted direction was consistent with the simulated heading preference of the stimulated sites (i.e., opposite to the preferred flow direction), and vice versa, for the sites with simulated heading preference opposite to the shifted direction. Across 87 sessions in the first monkey (Fig. 3C) and 67 sessions in the second monkey (Fig. 3D), nearly half of the cases (M1: n = 40, 56.3%; M2: n = 24, 44.4%) showed significant biases (P < 0.05, probit regression) and the majority of them (M1:38/40 = 95%; M2:22/24 = 91.7%) were predictable from the tuning curves, that is, opposite to the stimulated neurons’ preferred flow motion direction. On average, the mean PSE shift induced by microstimulation was significantly greater than 0 for both animals (M1:0.80° ± 0.10°, P = 2.8E-11; M2:0.81° ± 0.14°, P = 3.2E-7, two tailed t-test). This effect was similar, yet slightly weaker than that previously reported in area MSTd (~1.2°) under similar experimental parameters (Gu et al. 2012). However, the microstimulation effect in MT was not uniform across the population. We found that stimulating neurons with facilitated surrounds tended to produce larger PSE shift than stimulating neurons with inhibitory surrounds, based on 2 analyses. First, when plotting all data points of the induced PSE shift (including those insignificant ones) as a function of the SI in a continuous format, there was a significant positive trend (R = 0.20, P = 0.022, Spearman rank correlation, Fig. 4A). Second, we separated all data points into 2 categories according to the SI criterion, leading to one group favoring facilitated surrounds (red symbols in Fig. 4B), and the other group favoring inhibitory surrounds (purple symbols in Fig. 4B). We found both groups showed significant PSE shift (P < 0.05, t-test) with the former group tended to exhibit larger effects (Fig. 4B). Figure 4. View largeDownload slide Correlation between microstimulation effect and the surround properties of MT. (A) Induced PSE shift plotted as a function of the surround index. Solid line is the type II linear regression. Horizontal dashed line: no PSE shift. Filled and open symbols represent significant and insignificant PSE shifts as before. This analysis was conducted on 126 stimulated sites which were also tested the surround property during the experiment. Circle symbol: monkey R, triangle: monkey Y. (B) Mean induced PSE shift for 2 groups of MT units: neurons with more facilitated surrounds (red symbols) and with more inhibitory surrounds (purple symbols). The 2 groups of neurons were defined according to SI value explored under a broad range (from −0.1 to 0) rather than just 0. In such a case, there could be similar numbers of cells in each group. For both groups, the mean induced PSE shift was significantly > 0 for all data (t-test). In the gray area, the mean PSE shift from the SI > SIcriterion group was significantly larger than from the SI < SIcriterion group (P < 0.05, t-test). (C) Tuning similarity between the stimulated sites and the neighboring sites as a function of the surround property. Tuning similarity was measured by the correlation coefficient of the 2 tuning curves. Filled symbols: P < 0.05, open symbols: P > 0.05. Solid line indicates linear regression fit. Dashed line indicates correlation coefficient = 0. Figure 4. View largeDownload slide Correlation between microstimulation effect and the surround properties of MT. (A) Induced PSE shift plotted as a function of the surround index. Solid line is the type II linear regression. Horizontal dashed line: no PSE shift. Filled and open symbols represent significant and insignificant PSE shifts as before. This analysis was conducted on 126 stimulated sites which were also tested the surround property during the experiment. Circle symbol: monkey R, triangle: monkey Y. (B) Mean induced PSE shift for 2 groups of MT units: neurons with more facilitated surrounds (red symbols) and with more inhibitory surrounds (purple symbols). The 2 groups of neurons were defined according to SI value explored under a broad range (from −0.1 to 0) rather than just 0. In such a case, there could be similar numbers of cells in each group. For both groups, the mean induced PSE shift was significantly > 0 for all data (t-test). In the gray area, the mean PSE shift from the SI > SIcriterion group was significantly larger than from the SI < SIcriterion group (P < 0.05, t-test). (C) Tuning similarity between the stimulated sites and the neighboring sites as a function of the surround property. Tuning similarity was measured by the correlation coefficient of the 2 tuning curves. Filled symbols: P < 0.05, open symbols: P > 0.05. Solid line indicates linear regression fit. Dashed line indicates correlation coefficient = 0. Previous studies have shown that the magnitude of microstimulation effects tend to be related with 2 factors: clustering structure and tuning strength (Gu et al. 2012). In our experiment, most MT neurons around the microstimulation sites showed consistent surround properties  (see Supplementary Fig. 4), indicating either group of cells was well clustered. We then compared their strength of clustering by computing the correlation coefficients of the tuning curves between 2 neighboring sites. The average correlation coefficients were not significantly dependent on the surround property (R = −0.16, P = 0.16, Spearman rank correlations, Fig. 4C), indicating that the 2 groups of MT neurons shared similar clustering structure. Thus the relatively larger microstimulation effect observed from the cells with excitatory surround is unlikely due to the clustering factor. Instead, it may be related with the other factor: tuning strength (Fig. 2C; Supplementary Fig. 2C, D). Previous studies have shown that sensory neurons more sensitive to stimuli varied around the reference tended to contribute more to the animals’ behavioral choice (Purushothaman and Bradley 2005; Gu et al. 2012). In the current study, we also found that electrically stimulating MT neurons towards lateral motion preference generated slightly larger effects than neurons with other preferences (Supplementary Fig. 6). This was potentially because the former group contained more information about heading varied around straight ahead (Gu et al. 2010). To further test this, we estimated the precision of heading discrimination by computing Fisher information (Gu et al. 2010). Indeed, overall MT population showed comparable Fisher information as in area MSTd (Supplementary Fig. 2G). Within MT, neurons with excitatory surround were slightly more sensitive to heading stimuli than neurons with inhibitory surround. Hence, our results indicate that both MT populations with excitatory surround and with inhibitory surround are involved in the brain's heading computation, with the former group showing relatively larger contribution. To explore whether the microstimulation effect was robust under a broader range of the electrical current intensities, we tried different amplitudes peaked at 10, 40, 60 and even as large as 200 μA in addition to 20 μA used in the above described experiment (Fig. 5). For one example, the induced PSE shift was getting larger, while the thresholds were not much changed as the peak current amplitude was increased from 10 to 60 μA (Fig. 5A), potentially due to a larger proportion of neurons activated by the spread of the currents. However, under 200 uA, psychometric shifts were inconsistent and the function became flatter as evidenced by increased thresholds (Δthreshold = 2.8°, P = 0.027, probit test, Fig. 5B). This phenomenon suggested that too large electrical currents would activate too many neurons that produced inconsistent signals, that is,  noise into the neural circuits underlying heading decision, and consequently impaired the animals’ performance (Murasugi et al. 1993). This observation was also clearly reflected in the population (Fig. 5C–D): the induced PSE shift, rather than the threshold change, was monotonically increased as a function of the currents amplitude within 60μA (mean PSE shift: 10 μA: 0.40°, P = 0.25E-3; 20 μA: 0.98°, P = 5.6E-6; 40 μA: 1.13°, P = 2.8E-9; 60 μA: 1.64°, P = 1.82E-10, two tailed t-test; Mean threshold change: 10 μA: −0.04°, P = 0.76; 20 μA: 0.48°, P = 0.0004; 40 μA: 0.09°, P = 0.57; 60 μA: 0.55°, P = 0.0022, two tailed t-test). In contrast, the mean induced PSE shift was not significantly different from zero and its sign was in the wrong direction under 200μA condition (Mean PSE shift: −1.76°, P = 0.2, t-test). Instead, the mean threshold was substantially increased by 2.2° by the large currents (P = 0.0005, two tailed t-test), potentially due to currents spread into more distant areas influencing neurons with heterogeneous tuning properties. Thus, the weak currents (≤60 μA) used in the above experiments may have activated neurons within only a few hundreds of micrometers of the stimulated sites. More importantly, these data from both small and large currents demonstrated that MT not only had a sufficient, but also had a necessary role in the neural circuit responsible for heading computation. Figure 5. View largeDownload slide Microstimulation effects with a broader range of electrical current intensities. (A) Psychometric functions plotting the proportion of rightward responses as a function of heading direction and current intensity as the parameter: 0 uA (i.e., control, black), 10 uA (pink), 20 uA (red), 40 uA (blue), and 60 uA (green). (B) Psychometric functions for current intensity of 0 uA (i.e., control, solid black) and 200 uA (dotted blue). (C) Induced PSE shift as a function of current intensity for all neurons tested (10–60 μA: n = 41; 200 μA: n = 18). The black solid curve represents the average response. Positive values indicate a shift in direction consistent with the preferred direction of the stimulated neurons. (D) Averaged induced threshold change across the population of neurons (10–60 μA: n = 41; 200 μA: n = 18). Positive values indicate an increase in threshold with current intensity. Error bars: SEM. *0.01 < P < 0.05; **0.001 < P < 0.01; ***P < 0.001. Figure 5. View largeDownload slide Microstimulation effects with a broader range of electrical current intensities. (A) Psychometric functions plotting the proportion of rightward responses as a function of heading direction and current intensity as the parameter: 0 uA (i.e., control, black), 10 uA (pink), 20 uA (red), 40 uA (blue), and 60 uA (green). (B) Psychometric functions for current intensity of 0 uA (i.e., control, solid black) and 200 uA (dotted blue). (C) Induced PSE shift as a function of current intensity for all neurons tested (10–60 μA: n = 41; 200 μA: n = 18). The black solid curve represents the average response. Positive values indicate a shift in direction consistent with the preferred direction of the stimulated neurons. (D) Averaged induced threshold change across the population of neurons (10–60 μA: n = 41; 200 μA: n = 18). Positive values indicate an increase in threshold with current intensity. Error bars: SEM. *0.01 < P < 0.05; **0.001 < P < 0.01; ***P < 0.001. Microstimulation Effects without Visual Cues Present Inside MT Receptive Fields Given that MT units respond to only a small part of the visual field, it is unclear how MT microstimulation would affect the perception of global optic flow. One intuitive thought is that artificially activating MT neurons might produce distorted flow inside the receptive fields of the stimulated units, and subsequently biases heading judgments just like the local perturbation effects observed in the behavioral experiments. To test this possibility, we occluded the visual stimuli within the receptive fields of the electrical-stimulated MT neurons with an opaque, black round mask as large as 20–40° (Average: ~28°) in diameter. This size was chosen to be around 5 folds on average larger than the MT receptive field size (~6° on average in this experiment) in the following experiment to guarantee a complete block of the visual stimuli “viewed” by the isolated neurons (Fig. 6A, Supplementary Fig. 7A). As a result, neuronal responses were greatly diminished as shown in one typical example in Figure 6B (no mask, green curve: p = 7E-13; with mask, blue curve: P = 0.92, one-way ANOVA). We then applied microstimulation (20 μA) in half of the trials in both conditions, with and without masks. In this example, microstimulation evoked a significant PSE shift by 2.74° (P = 2.9E-5, probit regression) under the no-mask condition (Fig. 6D). Interestingly, under the mask condition, microstimulation also produced a significant PSE shift of 3.45° (P = 7.9E-7, probit regression, Fig. 6E), even though a large proportion of the electrical-stimulated MT neurons actually could not “see” (did not have access to) the visual stimuli within their receptive fields. Figure 6. View large Download slide Microstimulation effects with masks excluding the flow field within the receptive field. (A) Schematic illustration of the no-mask condition (left panel) and the mask condition (right panel). The black square illustrates the location of the mask. The dashed red circle illustrates the location of the mapped receptive field. The size of the mask is always slightly larger than the size of the receptive field so as to completely exclude any visual stimulation within the receptive fields. (B) Example of neuronal response as a function of heading direction for the no-mask condition (green curve) and mask condition (blue curve). Error bars: SEM. (C) Population neuronal response plotted as a function of angle relative to the preferred heading direction for the no-mask condition (green curve) and mask condition (blue curve) (M1: n = 15; M2: n = 13). Responses normalized to the maximum response. Error bars: SEM. (D, E) Psychometric functions showing the proportion of rightward responses are plotted as a function of heading direction, when the mask was present (D) or absent (E). Black: no electrical stimulation. Red: electrically stimulated trials. (F) Comparison between induced PSE shifts with (y-axis) and without mask (x-axis). Arrows indicate the mean values. Filled bars: P < 0.05; open bars: P > 0.05, probit regression. The asterisk represented for the example electrical stimulation case in (D, E). (G) Statistics for 13 sites from the population under no mask, mask, and blank conditions. In the blank condition, the mask covered the whole screen (except for the fixation point). For this condition, the psychometric functions are nearly flat and cannot be fit to reliably extract PSE (see Supplementary Fig. 5C). So we compute the difference in the proportion of responses towards the preferred direction of the stimulated neurons, for trials with and without electrical microstimulation. Error bars: SEM. *0.01 < P < 0.05; **0.001 < P < 0.01; ***P < 0.001. Figure 6. View large Download slide Microstimulation effects with masks excluding the flow field within the receptive field. (A) Schematic illustration of the no-mask condition (left panel) and the mask condition (right panel). The black square illustrates the location of the mask. The dashed red circle illustrates the location of the mapped receptive field. The size of the mask is always slightly larger than the size of the receptive field so as to completely exclude any visual stimulation within the receptive fields. (B) Example of neuronal response as a function of heading direction for the no-mask condition (green curve) and mask condition (blue curve). Error bars: SEM. (C) Population neuronal response plotted as a function of angle relative to the preferred heading direction for the no-mask condition (green curve) and mask condition (blue curve) (M1: n = 15; M2: n = 13). Responses normalized to the maximum response. Error bars: SEM. (D, E) Psychometric functions showing the proportion of rightward responses are plotted as a function of heading direction, when the mask was present (D) or absent (E). Black: no electrical stimulation. Red: electrically stimulated trials. (F) Comparison between induced PSE shifts with (y-axis) and without mask (x-axis). Arrows indicate the mean values. Filled bars: P < 0.05; open bars: P > 0.05, probit regression. The asterisk represented for the example electrical stimulation case in (D, E). (G) Statistics for 13 sites from the population under no mask, mask, and blank conditions. In the blank condition, the mask covered the whole screen (except for the fixation point). For this condition, the psychometric functions are nearly flat and cannot be fit to reliably extract PSE (see Supplementary Fig. 5C). So we compute the difference in the proportion of responses towards the preferred direction of the stimulated neurons, for trials with and without electrical microstimulation. Error bars: SEM. *0.01 < P < 0.05; **0.001 < P < 0.01; ***P < 0.001. We conducted this experiment in 28 sites on 2 monkeys. Across the population, a mask indeed efficiently blocked the visual responses of the neurons that were applied microstimulation in the later experiment (Fig. 6C, no mask, green curve: P = 7.0E-13; with mask, blue curve: P = 0.9, one-way ANOVA). On the behavioral level, adding the mask slightly increased the psychophysical threshold, but did not generate systematic bias (see Supplementary Fig. 7). However, under the mask condition, microstimulation still induced a significant PSE shift (M1:1.05° ± 0.37°, P = 0.0123, M2:2.00° ± 0.65°, P = 0.01, t-test, Fig. 6E, F), and this magnitude was not smaller than, but rather comparable to that under the no-mask condition (M1: P = 0.17, M2: P = 0.022, paired t-test, Fig. 6F). Thus, the effect of microstimulation in MT neurons on heading estimation does not require visual cues to be present within their receptive fields. This is not to say though, that visual stimuli are not required at all. In fact, if the mask was set to be large enough to blank the whole visual display (blank-screen condition, Fig. 6G), microstimulation could not produce significant effects on the animals’ choice any more (−0.91% ± 0.26%, P = 0.7, t-test). Taken together, these results implied that microstimulation of MT produced an internal directional signal, and this net excitatory signal was further integrated with the other visually induced activities (stimuli outside the masked area of the stimulated neurons) by downstream areas to compute heading. Notice that such results were dramatically different from previous microstimulation experiments in which the animals made perceptual decisions based on visual stimuli that were only available within the electrically stimulated neurons’ receptive fields (Salzman et al. 1990, 1992) (see Discussion for details). Interactions Between Microstimulation and Local Visual Perturbation Signals As above, microstimulation in MT (Fig. 3) and visual local perturbation (Fig. 1) generated a consistent effect on the animals’ heading perception. Specifically, in the microstimulation case, heading judgments were shifted in the direction opposite to MT neurons’ preferred flow direction. Similarly, in the visual perturbation case, heading judgments were also shifted in the direction opposite to the added motion vector. Thus we expect these 2 different sources of signals to interact when provided together. In other words, the question is, can downstream areas successfully distinguish an internal source of signal (microstimulation) from an externally triggered signal (local-distorted flow) when reading out information from MT? To address this, we conducted experiments with 4 different conditions (Fig. 7A): 1) a normal heading discrimination task without electrical stimulation (i.e., microstimulation) and local visual perturbation; 2) heading with microstimulation perturbation only; 3) heading with local visual perturbation only; and 4) heading with simultaneous microstimulation plus local visual perturbation. Note that in the fourth condition, the sign of the 2 perturbation effects could be either opposite or congruent. All the trials in the 4 conditions were randomly interleaved in 1 block. Figure 7. View largeDownload slide Interactions between microstimulation and visual perturbation effects. (A) Illustration of 4 experimental conditions interleaved randomly within one block. Upper left: normal self-motion condition without visual perturbation; upper right: perturbation induced by electrical microstimulation; lower left: perturbation by local object motion; lower right: combined effect of simultaneously introducing external (visual perturbation) and internal (microstimulation) perturbations; the red dashed circle represents the receptive field of the recorded/electrically stimulated MU. (B, C) Psychometric functions plotting the proportion of rightward responses as a function of heading direction, when microstimulation and local perturbation caused opposite (B) or congruent (C) effects. (D) Comparison between the PSE shift induced by local visual perturbation (y-axis) and microstimulation (x-axis). Open triangles indicate the mean values. (E) Comparison between the PSE shifts induced by vector summation (ordinate) and combined local perturbation and microstimulation (abscissa). Black: opposite condition; Grey: congruent condition. The inset bar graph gives the average values for congruent and opposite conditions. R: real PSE shift in combined condition. VS: PSE shift predicted by vector summation. (F) Number of cases showing different gains from microstimulation for opposite (black) and congruent (grey) conditions. Figure 7. View largeDownload slide Interactions between microstimulation and visual perturbation effects. (A) Illustration of 4 experimental conditions interleaved randomly within one block. Upper left: normal self-motion condition without visual perturbation; upper right: perturbation induced by electrical microstimulation; lower left: perturbation by local object motion; lower right: combined effect of simultaneously introducing external (visual perturbation) and internal (microstimulation) perturbations; the red dashed circle represents the receptive field of the recorded/electrically stimulated MU. (B, C) Psychometric functions plotting the proportion of rightward responses as a function of heading direction, when microstimulation and local perturbation caused opposite (B) or congruent (C) effects. (D) Comparison between the PSE shift induced by local visual perturbation (y-axis) and microstimulation (x-axis). Open triangles indicate the mean values. (E) Comparison between the PSE shifts induced by vector summation (ordinate) and combined local perturbation and microstimulation (abscissa). Black: opposite condition; Grey: congruent condition. The inset bar graph gives the average values for congruent and opposite conditions. R: real PSE shift in combined condition. VS: PSE shift predicted by vector summation. (F) Number of cases showing different gains from microstimulation for opposite (black) and congruent (grey) conditions. Figure 7B showed an example when microstimulation and local visual perturbation produced opposite effects. Specifically, microstimulation significantly shifted the psychometric function to leftward (ΔPSE: −1.40°, P = 2.5E-4, probit regression, red curve), while local visual perturbation shifted the function to rightward (ΔPSE: 1.38°, P = 0.0052, probit regression, green curve). Interestingly, these 2 signals appeared to be canceled out during simultaneous microstimulation and local visual perturbation: the induced PSE shift was not significant anymore (ΔPSE: 0.29°, P = 0.52, probit regression, dashed blue curve). In contrast, in the other example when microstimulation and local visual perturbation produced congruent effects (microstimulation: ΔPSE = 1.70°, P = 5.3E-4; visual perturbation: ΔPSE = 1.04°, P = 0.0125, probit regression, Fig. 7C, red and green curves, respectively), simultaneous perturbations significantly boosted the effect on the animal's performance (ΔPSE = 3.62°, P = 2.7E-7, probit regression, Fig. 7C, dashed blue curve). Across all 34 experiments (opposite case: n = 19; congruent case: n = 15), the magnitude of the ΔPSE induced by the 2 methods on average was roughly the same (microstimulation: 0.83° ± 0.13°; local visual perturbation: 1.04° ± 0.11°, P = 0.034, paired t-test, Fig. 7D). To understand how the 2 different sources of signals interact with each other, we tried 2 approaches. Firstly, we examined how well the induced PSE shift could be predicted by a straight summation of the ΔPSE measured from each source when both microstimulation and visual perturbation were simultaneously present (Fig. 7E). In fact, the 2 metrics were highly correlated (R = 0.79, P = 3.2E-8, Spearman rank correlation), and their means were not significantly different from each other (congruent case: 1.81° ± 0.28° vs. 1.20° ± 0.35°, P = 0.29, paired t-test; opposite case: 0.13° ± 0.13° vs. −0.06° ± 0.12°, P = 0.23, paired t-test, inset in Fig. 7E). Secondly, we computed a “microstimulation gain,” defined as the fraction of the ΔPSE induced by microstimulation that must be added to the visually induced PSE shift to explain the combined psychometric functions. If this gain equals 1, the effects of microstimulation and visual perturbation are the same. In contrast, a gain < 1 implies that the microstimulation makes a smaller contribution, and vice versa for a gain > 1. It turned out that the microstimulation gain amounted on average to a value of close to 1 for both opposite (1.05, P = 0.85, two tailed t-test, black bars in Fig. 7F) and congruent cases (0.73, P = 0.48, two tailed t-test, gray bars in Fig. 7F). Thus, the contribution of the microstimulation appeared to be analogous to that of the visual perturbation when they were applied together. These results implied that it is likely that microstimulation may have evoked similar responses as those from the visual cue, and the downstream areas in the brain such as MSTd were unlikely to distinguish whether the pooled MT signals were activated from an internal or external source. Discussion In the current study, we showed that MT neurons, containing much smaller receptive fields compared with their downstream neurons such as MSTd, were highly responsive to the global optic flow stimuli that were typically experienced during self-motion in natural navigation. Introducing a visual perturbation cue within a very small part of the visual field (<1% of the whole field) could significantly bias heading judgments in both human and monkeys based on the global optic flow cues, and shift the macaque MT tuning as well. Microstimulation experiments confirmed such a relationship is causal, both sufficient and necessary, indicating that MT plays a critical role in the neural circuit that is responsible for heading computation. In addition, our microstimulation experiment also revealed that the readout from MT neurons was not homogeneous, but was related to their surround effects. Such a weighted-pooling mechanism may mediate the biased effects of local perturbation cues on heading judgments as observed in many psychophysical studies (Warren and Saunders 1995; Royden and Hildreth 1996; Dokka et al. 2015; Layton and Fajen 2016). We thus infer that if the brain by any chance could use other sensory modality signals, for example, vestibular to release the influence on heading estimation from the perturbation signals, for example, independently moving objects (Dokka et al. 2015), this process may happen in the downstream areas that are multisensory, for example, MSTd/VIP (Gu et al. 2006,2008; Chen et al. 2011, 2013; Kim et al. 2016), but less likely in a visual-dominant area MT (Chowdhury et al. 2009). Behavioral Effects of Local-Distorted Flow on Heading Perception An impact of locally distorted flow fields simulating independently moving objects on heading estimation has been reported in several psychophysical studies (Warren and Saunders 1995; Royden and Hildreth 1996; Dokka et al. 2015; Layton and Fajen 2016). All of these studies have observed significant effects similar to ours, particularly when the object motion trajectory was across the FOE area. Such results support a weighted spatial pooling mechanism by which the brain integrates all local motion vectors across the whole visual field to compute heading (Warren and Saunders 1995). Even if we have tried several ways to boost the salience of the distorted flow field relative to the background flow, including the color, edge from the motion, dots density and relative motion speed, these manipulations could not help the subjects remove the influence of local perturbation on their heading judgment. In addition, we have instructed the human subjects to intentionally ignore the perturbed area, and rewarded the monkeys according to the global optic flow instead of the local perturbation signals. However, the bias effect still holds across the training process. Notice that this is not to say that the impact from object motion on self-motion perception cannot be modulated at all. For example, a recent study suggested that employing other sensory modalities such as vestibular inertial motion, can reduce this impact to some extent (Dokka et al. 2015), and this process may be mediated by multisensory neurons in the brain such as area MSTd (Kim et al. 2016). These results have important implications in that during natural navigation, heading estimation might be more robust when all signals from sensory modalities are available, compared with the situation when only visual cues are present. Functional Roles of MT in Local and Global Visual Processing MT has been traditionally defined in the dorsal visual stream (Dubner and Zeki 1971; Maunsell and Van Essen 1987; Britten 2008). MT neurons are highly sensitive to visual motion signals within their receptive fields (Tanaka et al. 1986; Maunsell and Van Essen 1987). Previous works have indicated that MT pools and integrates visual motion information from upstream areas including the primary visual cortex (V1) to compute a net motion vector (Britten and Heuer 1999), or a global plaid pattern (Pack and Born 2001; Born and Bradley 2005; Rust et al. 2006). However, all these stimuli were restricted inside MT's receptive fields that were much smaller compared with the whole visual field during natural navigation. Thus, motion signals in MT for each local spatial field need to be further integrated by downstream neurons (e.g., MSTd) for global heading computation (Supplementary Fig. 9). MT neurons are not homogeneous. A number of works have suggested that MT neurons could be separated into 2 groups according to their surround properties: neurons with antagonistic surrounds (MT−) and neurons with excitatory surrounds (MT+) (owl monkeys: (Allman et al. 1985; Born and Tootell 1992; Berezovskii and Born 2000; Born 2000); macaques: (Tanaka et al. 1986; Raiguel et al. 1995)). It has been speculated that the antagonistic group may be more involved in object-ground segregation and the facilitated group more involved in wide field stimuli processing (Tanaka et al. 1986; Born and Tootell 1992; Berezovskii and Born 2000; Born 2000; Orban 2008). In the current study, we also discovered that MT cells’ receptive fields carried surrounds with more or less inhibitory effect to the optic flow stimuli, as quantified by the SI. We noticed that the SI distribution of our data was more uniform and clustered around 0 (i.e., neither inhibition nor excitation), compared with the data reported previously (Born and Tootell 1992; Born 2000). We think that a number of factors may underlie this phenomenon. First, in previous work, visual stimuli were located exclusively in the laminar plane. Consequently, this maximized modulatory surround effects because inside and outside of the receptive fields were activated by dots of identical size and speed (Allman et al. 1985; Born and Bradley 2005). In our study, however, we used complex optic flow that contained motion parallax and size information. The surround effects may have been reduced by these heterogeneous stimuli. Second, there might be a species difference. In fact, the 2 categories of MT neurons have been reported to be much more clearly separated in owl monkeys compared with those in macaques (Tanaka et al. 1986; Born and Tootell 1992). Although the surround effects were more uniform in our case, we still observed a significant trend in that neurons with more excitatory surround tended to contribute more to the subject's heading estimation than cells with more inhibitory surrounds, as revealed by electrical microstimulation. To our knowledge, the only causal experiment showing separate functions of MT neurons is from Born's study, in which microstimulation on MT− and MT+ neurons produced opposite effects on the animal's smooth pursuit direction of a target moving in a direction opposite the background (Born et al. 2000). While these oculomotor data have been used to imply MT neurons are involved in segregation of local object and background motion (Born et al. 2000), our current results directly reveal the contributions from MT, particularly those units with excitatory surrounds in a fine heading discrimination task. Note, however, for the majority of MT cells with inhibitory surrounds, we also observed significant microstimulation effects, just with relatively smaller magnitude. As we have shown that MT neurons with different surround properties shared similar clustering structure (Supplementary Fig. 4), the heterogeneous microstimulation effects in the 2 groups of neurons (i.e., excitatory and inhibitory surround) observed in the current experiment should not be due to the clustering factor. Rather, these data indicate that heading computation based on global optic flow may be built on the weighted pooling of visual motion signals in both types of MT cells, explaining why a local perturbation signal can significantly bias the heading tuning curves of MSTd neurons (Logan and Duffy 2006; Also Supplementary Fig.10) as well as the subject's heading estimation as observed in the behavior (Warren and Saunders 1995; Dokka et al. 2015; Layton and Fajen 2016). Comparison with Previous Microstimulation Results Experiments using electrical microstimulation have been conducted in MT before to confirm that MT signals are sufficient for eliciting visual motion perception (Salzman et al. 1990, 1992; Murasugi et al. 1993; Nichols and Newsome 2002). The key difference in our current study is that, subjects need to judge their heading directions based on global optic flow (Britten and van Wezel 1998; Gu et al. 2008, 2012), rather than local motion information restricted to the stimulated neurons’ receptive field as used in the previous studies. This different paradigm may explain why our microstimulation effects persist during the mask condition, while the effects were absent if the visual stimuli were removed from the receptive fields of MT neurons during stimulus duration period as in previous studies (Salzman et al. 1990, 1992; Murasugi et al. 1993). Specifically, in our task, the brain pools local spatial information across the whole visual field to estimate heading. Thus although the visual stimuli were excluded from the stimulated MT neurons, the brain still integrated the internally triggered signals (i.e., microstimulation) with those elicited by external signals (i.e., visual) from the other MT neurons to compute heading. However, increasing the masked area to cover the whole visual display effectively removed the effects of microstimulation. This result may indicate a sensory “gating” mechanism (Seidemann et al. 1998) for heading in the downstream areas. For example, MSTd pools information from upstream MT no matter whether the responses are evoked from artificial electrical stimulation or visual stimuli. However, this pooling process is “gated” if no visual information is provided at all as in the blank-screen condition. Finally, it remains an open question what exactly the monkeys perceive within the microstimulated receptive fields in the absence of visual stimulation. This issue probably can only be solved by testing human subjects (Cohen and Newsome 2004; Reppas and Newsome 2007), as those studies conducted in patients (Rauschecker et al. 2011; Becker et al. 2013). Our paradigm using global optic flow to test the impact of microstimulation in MT also made it possible to qualitatively compare the results to MSTd results as reported in previous works (Britten and van Wezel 1998, 2002; Gu et al. 2012). It turned out that the microstimulation effects on the monkey's heading estimation were analogous in both areas, suggesting that the directional information useful for further heading computation in MT is largely preserved when propagated into MSTd. Taken together, the causal results from our current work in MT, as well as those from previous experiments in MSTd clearly demonstrate that the MT-MSTd circuit in the dorsal visual pathway play a key role (both sufficient and necessary) in global heading computation. Notice though, works from Britten's group also indicated that the microstimulation effects in MSTd became more salient if smooth eye pursuits were accompanied at the same time during heading. Whether this is also true in area MT requires further research in the future. Supplementary Material Supplementary material is available at Cerebral Cortex online. Funding Grants from the National Natural Science Foundation of China Project (31471048), the Strategic Priority Research Program of CAS (XDB02010000), the National Key Basic Research Project (2016YFC1306801), and the Shanghai Key Basic Research Project (16JC1420201). Notes We thank Wenyao Chen for monkey care and training; Ying Liu for software programming. Conflict of Interest: None declared. References Allman J, Miezin F, McGuinness E. 1985. Direction- and velocity-specific responses from beyond the classical receptive field in the middle temporal visual area (MT). Perception . 14: 105– 126. Google Scholar CrossRef Search ADS PubMed  Allman JM, Kaas JH. 1971. A representation of the visual field in the caudal third of the middle tempral gyrus of the owl monkey (Aotus trivirgatus). Brain Res . 31: 85– 105. Google Scholar CrossRef Search ADS PubMed  Becker HG, Haarmeier T, Tatagiba M, Gharabaghi A. 2013. Electrical stimulation of the human homolog of the medial superior temporal area induces visual motion blindness. J Neurosci . 33: 18288– 18297. Google Scholar CrossRef Search ADS PubMed  Berezovskii VK, Born RT. 2000. Specificity of projections from wide-field and local motion-processing regions within the middle temporal visual area of the owl monkey. J Neurosci . 20: 1157– 1169. Google Scholar PubMed  Born RT. 2000. Center-surround interactions in the middle temporal visual area of the owl monkey. J Neurophysiol . 84: 2658– 2669. Google Scholar CrossRef Search ADS PubMed  Born RT, Tootell RB. 1991. Single-unit and 2-deoxyglucose studies of side inhibition in macaque striate cortex. Proc Natl Acad Sci USA . 88: 7071– 7075. Google Scholar CrossRef Search ADS PubMed  Born RT, Tootell RB. 1992. Segregation of global and local motion processing in primate middle temporal visual area. Nature . 357: 497– 499. Google Scholar CrossRef Search ADS PubMed  Born RT, Bradley DC. 2005. Structure and function of visual area MT. Annu Rev Neurosci . 28: 157– 189. Google Scholar CrossRef Search ADS PubMed  Born RT, Groh JM, Zhao R, Lukasewycz SJ. 2000. Segregation of object and background motion in visual area MT: effects of microstimulation on eye movements. Neuron . 26: 725– 734. Google Scholar CrossRef Search ADS PubMed  Britten KH. 2008. Mechanisms of self-motion perception. Annu Rev Neurosci . 31: 389– 410. Google Scholar CrossRef Search ADS PubMed  Britten KH, van Wezel RJ. 1998. Electrical microstimulation of cortical area MST biases heading perception in monkeys. Nat Neurosci . 1: 59– 63. Google Scholar CrossRef Search ADS PubMed  Britten KH, Heuer HW. 1999. Spatial summation in the receptive fields of MT neurons. J Neurosci . 19: 5074– 5084. Google Scholar PubMed  Britten KH, Van Wezel RJ. 2002. Area MST and heading perception in macaque monkeys. Cereb Cortex . 12: 692– 701. Google Scholar CrossRef Search ADS PubMed  Chen A, DeAngelis GC, Angelaki DE. 2011. Representation of vestibular and visual cues to self-motion in ventral intraparietal cortex. J Neurosci . 31: 12036– 12052. Google Scholar CrossRef Search ADS PubMed  Chen A, DeAngelis GC, Angelaki DE. 2013. Functional specializations of the ventral intraparietal area for multisensory heading discrimination. J Neurosci . 33: 3567– 3581. Google Scholar CrossRef Search ADS PubMed  Chowdhury SA, Takahashi K, DeAngelis GC, Angelaki DE. 2009. Does the middle temporal area carry vestibular signals related to self-motion. J Neurosci . 29: 12020– 12030. Google Scholar CrossRef Search ADS PubMed  Cohen MR, Newsome WT. 2004. What electrical microstimulation has revealed about the neural basis of cognition. Curr Opin Neurobiol . 14: 169– 177. Google Scholar CrossRef Search ADS PubMed  Desimone R, Ungerleider LG. 1986. Multiple visual areas in the caudal superior temporal sulcus of the macaque. J Comp Neurol . 248: 164– 189. Google Scholar CrossRef Search ADS PubMed  Dokka K, DeAngelis GC, Angelaki DE. 2015. Multisensory integration of visual and vestibular signals improves heading discrimination in the presence of a moving object. J Neurosci . 35: 13599– 13607. Google Scholar CrossRef Search ADS PubMed  Dubner R, Zeki SM. 1971. Response properties and receptive fields of cells in an anatomically defined region of the superior temporal sulcus in the monkey. Brain Res . 35: 528– 532. Google Scholar CrossRef Search ADS PubMed  Duffy CJ, Wurtz RH. 1993. An illusory transformation of optic flow fields. Vision Res . 33: 1481– 1490. Google Scholar CrossRef Search ADS PubMed  Fetsch CR, Pouget A, DeAngelis GC, Angelaki DE. 2011. Neural correlates of reliability-based cue weighting during multisensory integration. Nat Neurosci . 15: 146– 154. Google Scholar CrossRef Search ADS PubMed  Gibson JJ. 1950. Perception of the visual world . Boston: Houghton-Mifflin. Gu Y, Angelaki DE, DeAngelis GC. 2008. Neural correlates of multisensory cue integration in macaque MSTd. Nat Neurosci . 11: 1201– 1210. Google Scholar CrossRef Search ADS PubMed  Gu Y, DeAngelis GC, Angelaki DE. 2012. Causal links between dorsal medial superior temporal area neurons and multisensory heading perception. J Neurosci . 32: 2299– 2313. Google Scholar CrossRef Search ADS PubMed  Gu Y, Watkins PV, Angelaki DE, DeAngelis GC. 2006. Visual and nonvisual contributions to three-dimensional heading selectivity in the medial superior temporal area. J Neurosci . 26: 73– 85. Google Scholar CrossRef Search ADS PubMed  Gu Y, Fetsch CR, Adeyemo B, DeAngelis GC, Angelaki DE. 2010. Decoding of MSTd population activity accounts for variations in the precision of heading perception. Neuron . 66: 596– 609. Google Scholar CrossRef Search ADS PubMed  Kim HR, Pitkow X, Angelaki DE, DeAngelis GC. 2016. A simple approach to ignoring irrelevant variables by population decoding based on multisensory neurons. J Neurophysiol . 116: 1449– 1467. Google Scholar CrossRef Search ADS PubMed  Komatsu H, Wurtz RH. 1988. Relation of cortical areas MT and MST to pursuit eye movements. I. Localization and visual properties of neurons. J Neurophysiol . 60: 580– 603. Google Scholar CrossRef Search ADS PubMed  Layton OW, Fajen BR. 2016. Sources of bias in the perception of heading in the presence of moving objects: object-based and border-based discrepancies. J Vis . 16: 9. Google Scholar CrossRef Search ADS PubMed  Layton OW, Mingolla E, Browning NA. 2012. A motion pooling model of visually guided navigation explains human behavior in the presence of independently moving objects. J Vis . 12: 1– 19. Logan DJ, Duffy CJ. 2006. Cortical area MSTd combines visual cues to represent 3-D self-movement. Cereb Cortex . 16: 1494– 1507. Google Scholar CrossRef Search ADS PubMed  Maunsell JH, Newsome WT. 1987. Visual processing in monkey extrastriate cortex. Annu Rev Neurosci . 10: 363– 401. Google Scholar CrossRef Search ADS PubMed  Maunsell JH, Van Essen DC. 1987. Topographic organization of the middle temporal visual area in the macaque monkey: representational biases and the relationship to callosal connections and myeloarchitectonic boundaries. J Comp Neurol . 266: 535– 555. Google Scholar CrossRef Search ADS PubMed  Murasugi CM, Salzman CD, Newsome WT. 1993. Microstimulation in visual area MT: effects of varying pulse amplitude and frequency. J Neurosci . 13: 1719– 1729. Google Scholar PubMed  Nichols MJ, Newsome WT. 2002. Middle temporal visual area microstimulation influences veridical judgments of motion direction. J Neurosci . 22: 9530– 9540. Google Scholar PubMed  Orban GA. 2008. Higher order visual processing in macaque extrastriate cortex. Physiol Rev . 88: 59– 89. Google Scholar CrossRef Search ADS PubMed  Pack CC, Born RT. 2001. Temporal dynamics of a neural solution to the aperture problem in visual area MT of macaque brain. Nature . 409: 1040– 1042. Google Scholar CrossRef Search ADS PubMed  Prince SJ, Pointon AD, Cumming BG, Parker AJ. 2002. Quantitative analysis of the responses of V1 neurons to horizontal disparity in dynamic random-dot stereograms. J Neurophysiol . 87: 191– 208. Google Scholar CrossRef Search ADS PubMed  Purushothaman G, Bradley DC. 2005. Neural population code for fine perceptual decisions in area MT. Nat Neurosci . 8: 99– 106. Google Scholar CrossRef Search ADS PubMed  Raiguel S, Van Hulle MM, Xiao DK, Marcar VL, Orban GA. 1995. Shape and spatial distribution of receptive fields and antagonistic motion surrounds in the middle temporal area (V5) of the macaque. Eur J Neurosci . 7: 2064– 2082. Google Scholar CrossRef Search ADS PubMed  Rauschecker AM, Dastjerdi M, Weiner KS, Witthoft N, Chen J, Selimbeyoglu A, Parvizi J. 2011. Illusions of visual motion elicited by electrical stimulation of human MT complex. PLoS One . 6: e21798. Google Scholar CrossRef Search ADS PubMed  Reppas JB, Newsome WT. 2007. Brain stimulation: feeling the buzz. Curr Biol . 17: R358– R360. Google Scholar CrossRef Search ADS PubMed  Royden CS. 2002. Computing heading in the presence of moving objects: a model that uses motion-opponent operators. Vision Res . 42: 3043– 3058. Google Scholar CrossRef Search ADS PubMed  Royden CS, Hildreth EC. 1996. Human heading judgments in the presence of moving objects. Percept Psychophys . 58: 836– 856. Google Scholar CrossRef Search ADS PubMed  Rust NC, Mante V, Simoncelli EP, Movshon JA. 2006. How MT cells analyze the motion of visual patterns. Nat Neurosci . 9: 1421– 1431. Google Scholar CrossRef Search ADS PubMed  Salzman CD, Britten KH, Newsome WT. 1990. Cortical microstimulation influences perceptual judgements of motion direction. Nature . 346: 174– 177. Google Scholar CrossRef Search ADS PubMed  Salzman CD, Murasugi CM, Britten KH, Newsome WT. 1992. Microstimulation in visual area MT: effects on direction discrimination performance. J Neurosci . 12: 2331– 2355. Google Scholar PubMed  Seidemann E, Zohary E, Newsome WT. 1998. Temporal gating of neural signals during performance of a visual discrimination task. Nature . 394: 72– 75. Google Scholar CrossRef Search ADS PubMed  Tanaka K, Hikosaka K, Saito H, Yukie M, Fukada Y, Iwai E. 1986. Analysis of local and wide-field movements in the superior temporal visual areas of the macaque monkey. J Neurosci . 6: 134– 144. Google Scholar PubMed  Van Essen DC, Maunsell JH, Bixby JL. 1981. The middle temporal visual area in the macaque: myeloarchitecture, connections, functional properties and topographic organization. J Comp Neurol . 199: 293– 326. Google Scholar CrossRef Search ADS PubMed  Warren WH Jr., Saunders JA. 1995. Perceiving heading in the presence of moving objects. Perception . 24: 315– 331. Google Scholar CrossRef Search ADS PubMed  Warren WH Jr., Morris MW, Kalish M. 1988. Perception of translational heading from optical flow. J Exp Psychol Hum Percept Perform . 14: 646– 660. Google Scholar CrossRef Search ADS PubMed  © The Author 2017. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com
TI - Causal Evidence of Motion Signals in Macaque Middle Temporal Area Weighted-Pooled for Global Heading Perception
JF - Cerebral Cortex
DO - 10.1093/cercor/bhw402
DA - 2018-02-01
UR - https://www.deepdyve.com/lp/oxford-university-press/causal-evidence-of-motion-signals-in-macaque-middle-temporal-area-cMqus009Au
SP - 612
EP - 624
VL - 28
IS - 2
DP - DeepDyve
ER -