The visual motion detectors underlying ocular following responses in monkeys

Abstract

Psychophysical evidence indicates that visual motion can be sensed by low-level (energy-based) and high-level (feature-based) mechanisms. The present experiments were undertaken to determine which of these mechanisms mediates the initial ocular following response (OFR) that can be elicited at ultra-short latencies by sudden motion of large-field images. We used the methodology of Sheliga, Chen, FitzGibbon and Miles (Initial ocular following in humans: a response to first-order motion energy. Vision Research, In press), who studied the initial OFRs of humans, to study the initial OFRs of monkeys. Accordingly, we applied horizontal motion to 1) vertical square-wave gratings lacking the fundamental (“missing fundamental stimulus”), and 2) vertical grating patterns consisting of the sum of two sinusoids of frequency 3f and 4f, which created a repeating pattern with beat frequency, f. Both visual stimuli share a critical property: when subject to ¼-wavelength steps, their overall pattern (feature) shifts in the direction of the steps, whereas their major Fourier component shifts in the reverse direction (because of spatial aliasing). We found that the initial OFRs of monkeys to these stimuli, like those of humans, were always in the opposite direction to the ¼-wavelength shifts, i.e., in the direction of the major Fourier component, consistent with detection by (low-level) oriented spatio-temporal filters as in the well-known energy model of motion analysis. Our data indicate that the motion detectors mediating the initial OFR have quantitatively similar properties in monkeys and humans, suggesting that monkeys provide a good animal model for the human OFR.

1. Introduction

In primates, sudden movements of the visual scene elicit ocular following responses (OFRs) with ultra-short latencies: <80 ms in humans and <60 ms in monkeys with some stimuli (Gellman, Carl & Miles, 1990; Miles, Kawano & Optican, 1986; Sheliga, Chen, Fitzgibbon & Miles, 2005a; 2005b). It is believed that OFRs assist in the rapid stabilization of gaze with respect to the (stationary) surroundings (Miles, 1998), and there is substantial evidence that, in monkeys, it is mediated in large part by cortical area MST (Kawano, Inoue, Takemura, Kodaka & Miles, 2000; Kawano, Shidara, Watanabe & Yamane, 1994; Takemura, Inoue & Kawano, 2002). Because the OFR is driven by retinal image motion, it has recently been used to study some of the neural processes underlying the sensing of visual motion in humans (Chen, Sheliga, Fitzgibbon & Miles, 2005; Masson, Busettini, Yang & Miles, 2001; Masson & Castet, 2002; Masson, Yang & Miles, 2002; Sheliga et al., 2005a; 2005b; Yang & Miles, 2003). It is generally believed that there are at least two different neural mechanisms by which we analyze visual motion, and various descriptors have been applied to them, based in part on the methodology that was used to investigate them: see Lu and Sperling (2001) for review. One mechanism utilizes low-level visual motion detectors that are sensitive to luminance modulation, and functions without regard for form or features. This mechanism is variously referred to as, first-order, Fourier or energy-based. The other, high-level, mechanism(s) extracts visual motion using features to which the low-level visual motion detectors are insensitive, and is referred to as second-order, non-Fourier or feature-based. We will use these various terms interchangeably, although they are not strictly synonymous.

Using special broad-band stimuli, Sheliga et al. (2005a; 2005b) recently demonstrated that the initial OFR in humans is largely determined by the motion of the principal Fourier component of the visual stimulus, as though reliant on oriented spatiotemporal visual filters as in the well-known energy model of motion detection, i.e., the human OFR is mediated by low-level motion detectors. These workers elicited OFRs with two kinds of apparent-motion stimuli whose features and principal Fourier components moved in opposite directions under the assumption that the brain gives the greatest weight to the nearest-neighbor matches (Georgeson & Harris, 1990; Hammett, Ledgeway & Smith, 1993; see also Pantle & Turano, 1992). One was the missing fundamental (mf) stimulus, which can be constructed from a square wave by subtracting the fundamental sine-wave component. In the frequency domain, the mf stimulus consists of summed odd harmonics, the largest being the 3^rd and the remainder having progressively decreasing amplitudes such that the i^th harmonic has an amplitude proportional to 1/i. Sheliga et al (2005a; 2005b) moved the mf stimulus in discrete ¼-wavelength steps rather than smoothly because, when so moved, all of its harmonics are shifted ¼ of their wavelengths, the 4n+1 harmonics (where n is an integer) shifting in the direction of the actual image shifts (i.e., forwards, along with the entire pattern and its features), whereas the 4n−1 harmonics (which include the principal Fourier component, the 3^rd harmonic) shift in the opposite direction (i.e., backwards). Notice that a ¾-wavelength forward shift of a pure sine wave is exactly equivalent to a ¼-wavelength backward shift because of spatial aliasing, and it is invariably the latter that determines the direction of any associated OFRs and perceived motion (Pantle & Turano, 1992). The other apparent motion stimulus used by Sheliga et al was the so-called 3f4f stimulus, which is a repeating pattern with a spatial frequency of f that is constructed by summing two sinusoids of equal amplitude whose spatial frequencies are in the ratio 3:4, the 3f and 4f components (Hammett et al., 1993). When this pattern is shifted forwards in successive steps that are each ¼ of the wavelength of the beat, the 4f component is effectively stationary while the 3f component steps backward ¼ of its wavelength because of spatial aliasing, exactly as with the 4n−1 harmonics of the mf stimulus. Sheliga et al. (2005a; 2005b) reported that the initial OFRs elicited by these mf and 3f4f stimuli in humans were invariably in the direction of motion of their principal Fourier component(s) rather than the direction of motion of their features, consistent with a mechanism that has oriented spatio-temporal filters and senses 1^st-order motion (cf., Masson et al., 2002). Indeed, Sheliga et al (2005a) have suggested that the initial OFR of humans provides a model system for studying neural sensors that respond selectively to 1^st-order visual motion and do not respond to 2^nd-order motion. Psychophysical studies indicate that when ¼-wavelength steps are applied to mf stimuli human observers generally—but not always—perceive them to move in the opposite direction to their true motion (Adelson, 1982; Adelson & Bergen, 1985; Baro & Levinson, 1988; Brown & He, 2000; Georgeson & Harris, 1990; Georgeson & Shackleton, 1989). Also, human observers generally experience motion transparency when viewing 3f4f stimuli, seeing rapid forward motion of the beat and slower reverse motion of the 3f component, consistent with the simultaneous activation of feature-based (2^nd-order) and energy-based (1^st-order) sensing mechanisms, respectively (Hammett et al., 1993).

To further examine the neural mechanisms underlying the visual motion detectors underlying initial OFRs, it is critical to have an adequate animal model. We were interested in the monkey as one such model but were concerned by a previous report, based on one monkey, which concluded that the initial OFRs to motion defined by contrast modulated noise—a pure 2^nd-order motion stimulus—were essentially the same as to motion defined by luminance modulations—1^st-order motion—except for a very slight difference in latency (Benson & Guo, 1999). Further, there have been a number of reports that some neurons in the middle temporal area (MT) of the monkey are sensitive to 2^nd-order motion (Albright, 1992; Ilg & Churan, 2004; O’Keefe & Movshon, 1998). We now report that the initial OFRs elicited in macaque monkeys when ¼-wavelength steps are applied to mf and 3f4f stimuli are in all essentials like those described by Sheliga et al (2005a) in humans—being dominated in large part by the 1^st-order motion energy in the principal Fourier components—indicating that the monkey is a good animal model.

2. Methods

Data were collected from two rhesus monkeys (Macaca fuscata), weighing 7.7 (A) and 8.8 kg (B). All procedures reported here were approved by the Institute’s Animal Care and Use Committee. Many of the general procedures were the same as those used in previous studies of ocular tracking in monkeys (Kawano et al., 1994; Kodaka, Miura, Suehiro, Takemura & Kawano, 2004) and humans (Sheliga et al., 2005a) and so will only be given in brief.

2.1. Animal preparations

The monkeys were previously trained to fixate a small spot. Under pentobarbital sodium anesthesia and aseptic conditions, each monkey was implanted with a head holder, which allowed the head to be fixed in the standard stereotaxic position during the experiments, and a scleral search coil to allow measurement of the position of the right eye (Judge, Richmond & Chu, 1980).

2.2. Visual display and stimuli

The animals faced a 19 inch CRT monitor (Eizo T766, driven by a PC Radeon 9800 Pro video card), which was 50 cm in front of the eyes, in a dark room. Visual stimuli were presented on the monitor (resolution, 1280 × 1024 pixels; vertical refresh rate, 100 Hz). The RGB signals from the video card were converted to black and white images with 11-bit grayscale resolution through an attenuator (Pelli, 1997), exactly as described by Sheliga et al. (2005a). Briefly, a luminance look-up table with 256 equally-spaced luminance levels ranging from 0.3 cd/m² to 77.1 cd/m² was created by direct luminance measurements (LS-100 photometer; Konica-Minolta, Japan) under software control. This table was then expanded to 2048 equally-spaced levels by interpolation.

The visual images consisted of one-dimensional vertical grating patterns that could have one of four horizontal luminance profiles in any given trial: 1) a square wave with a missing fundamental (mf stimulus), achieved by summing the odd harmonics (starting with the 3^rd harmonic and finishing with the highest harmonic that fell short of the Nyquist frequency) as described by Sheliga et al (2005a); 2) a sum of two equal-amplitude sinusoids whose spatial frequencies were in the ratio 3:4 (3f4f stimulus); 3) a pure sine wave with the same frequency as the beat of the mf and 3f4f stimuli (1f stimulus); 4) a pure sine wave whose spatial frequency was three times that of the 1f stimulus (3f stimulus), and hence was the same as that of the principal Fourier component of the mf stimulus (the 3^rd harmonic) and the 3f component of the 3f4f stimulus. Each image extended 360 mm horizontally (39.6°; 1280 pixels) and 270 mm vertically (30.2°; 1024 pixels) and had a mean luminance of 38.7 cd/m². The initial phase of a given grating was randomized from trial to trial at intervals of ¼-wavelength. Motion was created by substituting a new image every frame (i.e., every 10 ms) for a total of 15 frames (i.e., stimulus duration, 150 ms), each new image being identical to the previous one except phase shifted horizontally by ¼ of the wavelength of the fundamental. In any given trial the successive steps were all in the same direction (rightward or leftward, randomly selected). We examined the OFRs elicited by these apparent motion stimuli at various stimulus contrast levels (1%, 2%, 4%, 8%, 16%, 32%, and 64%, where the contrast was defined as, ((L_max−L_min)/(L_max + L_min))*100%, L_max and L_min being the maximum and minimum luminance levels, respectively). An experimental block consisted of 56 stimulus entries (4 grating patterns, 2 directions of motion, 7 contrast levels), all of which were interleaved and randomly ordered in individual blocks.

2.3. Procedures

At the beginning of each trial, a grating pattern appeared together with a central target spot (diameter, 0.4°) that the animal had been trained to fixate. After the monkey’s right eye had been positioned within 2° of the fixation target for a randomized period of 750 to 1000 ms (and no saccades had been detected for the last 250 ms in this period), the fixation target disappeared and the apparent-motion stimulus began. Otherwise, the screen became uniform gray and the trial was repeated. The motion lasted for 150 ms, at which time the screen became a uniform gray with the same mean luminance. Then the animals were rewarded with a drop of juice, signaling the end of the trial. After an inter-trial interval of 1000–1500 ms, a new grating pattern appeared together with a fixation point, commencing a new trial. Data were collected over several sessions until each condition had been repeated an adequate number of times to permit good resolution of the responses through averaging.

2.4. Data collection and analyses

All aspects of the experimental paradigms were controlled by two PCs, which communicated via Ethernet using the TCP/IP protocol. One of the PCs was running a Real-time EXperimentation software package (REX) developed by Hays, Richmond and Optican (1982), and provided the overall control of the experimental protocol as well as acquiring, displaying, and storing the eye-movement data. The other PC was running Matlab subroutines, utilizing the Psychophysics Toolbox extensions (Brainard, 1997; Pelli, 1997), and generated the visual stimuli upon receiving a start signal from the REX machine.

Eye movements were measured using the electromagnetic search coil technique (Fuchs & Robinson, 1966). The voltage signals encoding the horizontal and vertical components of the eye position were passed through an analog low-pass filter (−3dB at 200 Hz) and were digitized to a resolution of 12 bits, sampling at 1kHz. All data were stored and transferred to another PC for analysis using computer programs based on Matlab (The Mathworks Inc.). The eye-position data were smoothed with a 3-pole digital butterworth filter (−3 dB at 30 Hz), and eye velocity traces were derived from the two-point backward difference. Eye acceleration profiles were derived from the two-point backward difference of the eye velocity traces, and were used to detect small saccades that went undetected during the experiment. Trials with saccadic intrusions were then discarded (on average, 23% in monkey A and 27% in monkey B). Mean temporal eye velocity profiles were computed for each of the stimulus conditions. To obtain low-noise estimates of eye velocity, responses were averaged over at least 31 trials in the preliminary experiment on spatial-frequency and 40 trials in the main experiment on contrast.

The initial horizontal OFRs were quantified by measuring the changes in horizontal eye position over the 50-ms time periods starting 50 ms after the onset of the motion stimuli. The minimum latency of onset was ~50 ms so that these response measures were restricted to the period prior to the closure of the visual feedback loop (i.e., twice the reaction time): initial open-loop responses. The responses to rightward and leftward were pooled to improve the signal-to-noise ratio by subtracting the mean response to a given leftward motion stimulus from the mean response to the corresponding rightward motion stimulus, and these will be referred to the “R-L responses”. The 95% confidence intervals were calculated to indicate the extent of the fluctuations in R-L measures. Because rightward eye movements were positive in our sign convention, these pooled R-L measures were positive when the OFR was in the direction of the applied image shift (also sometimes referred to as the “forward” direction, in contradistinction to the “backward” or “reverse” direction).

3. Results

3.1. Dependence on spatial frequency

In order to determine what spatial frequency to use for the mf and 3f4f stimuli, we did a preliminary experiment in which we applied ¼-wavelength steps to pure sinusoidal vertical gratings (contrast, 32%) with a wide range of spatial frequencies (0.05, 0.10, 0.20, 0.40, 0.81, 1.62 cycles/°). Figure 1 shows the data obtained from both monkeys, with the mean R-L velocity traces over time above (A, C) and the associated mean R-L change-in-position measures plotted against spatial frequency below (B, D). All OFRs were in the forward direction, which is positive in our convention (see Methods) and so they are rendered as upward deflections of the R-L velocity traces (from a zero baseline) in Fig. 1A, C. The R-L measures displayed clear band-pass dependence on spatial frequency that was well represented by Gaussian functions (r² values: 0.995, 0.996) with peaks (f_o) at 0.158, and 0.291 cycles/° and standard deviations (s) of 0.534 and 0.482 log units in monkey A and monkey B, respectively: see the continuous smooth curves in Fig. 1B,D and note the logarithmic abscissas. These parameters of the fitted Gaussian functions were used to derive a low-frequency cutoff (f_lo) and a high-frequency cutoff (f_hi), defined as the spatial frequencies at which the tuning curve was half its maximum, using the following expression from Read & Cumming (2003): $f_{o}exp(\pm \sigma \sqrt{ln4})$. The computed values of f_lo were 0.037, and 0.079 cycles/°, and the computed values of f_hi were 0.672, and 1.077 cycles/°. The initial OFRs of humans show a very similar Gaussian dependence on log spatial frequency (Sheliga et al., 2005a). Based on these findings, the fundamental spatial frequencies (f) selected for the mf and 3f4f stimuli were 0.09 cycles/° for monkey A and 0.15 cycles/° for monkey B, thereby ensuring that the initial OFRs elicited by pure sine waves of spatial frequency 1f and 3f were of similar amplitude.

Figure 1.

The initial OFR elicited by ¼-wavelength steps applied to sinusoidal grating patterns (1f stimulus): dependence on spatial frequency (two monkeys). A,C: Mean R-L eye velocity temporal profiles: see key for spatial frequencies (in cycles/°). Upward deflections of the traces from the zero baseline (dashed horizontal line) denote forward motion (in the direction of the steps). Zero on the abscissa denotes the time of the first ¼-wavelength shift (defined as the onset of stimulus motion). B,D: Mean R-L response measures (mean change in R-L position during the time period 50–100 ms after the onset of motion) plotted as a function of the spatial frequency in cycles/°; note that the abscissa has a logarithmic scale. The smooth curves are the best fit Gaussian functions: see Sheliga et al. (2005a) for methods. The data for monkey A are in A,B (31–37 trials per condition; SD’s ranged 0.018–0.072°) and the data for monkey B are in C,D (59–73 trials per condition; SD’s ranged 0.039–0.089°). Contrast, 32%. Error bars, 95% confidence intervals.

3.2. Dependence on contrast

Figure 2 shows the mean OFR temporal profiles (again, R-L responses) elicited from monkey A by successive discrete phase shifts applied to each of the four different grating patterns over a wide range of contrasts. The shifts always had the same absolute amplitude, 2.78°, which meant that with each shift the 1f, mf and 3f4f gratings stepped forwards ¼ of their wavelength (given that their wavelengths were all 11.1°), whereas the 3f grating stepped forwards ¾ of its wavelength, which was equivalent to a backward step of ¼ of its wavelength. Let us first consider the data obtained with the pure sine-wave stimuli. It is evident from Fig. 2 that the initial OFRs elicited by the 1f stimulus were always in the forward direction whereas those elicited by the 3f stimulus were always in the backward direction, exactly in accord with the shortest-path or nearest-neighbor matches for these stimuli: spatial aliasing. Of course, a pure sine wave has only one Fourier component and this always shifts together with its features (peaks and troughs), so it is not possible to determine which of these two attributes of the motion stimulus elicited the OFR here. Turning to the more complex grating patterns, however, it is evident that the initial OFR elicited by the mf and 3f4f stimuli were always in the backward direction, which was the direction of motion of their principal Fourier components (the 3^rd harmonic and the 3f component, respectively) and opposite to the direction of motion of their features.

Figure 2.

The initial OFR: dependence on contrast (eye velocity traces from monkey A). All motion stimuli consisted of successive shifts that had the same absolute amplitude, 2.78°, which was ¼ of the wavelength of the 1f, mf, and 3f4f gratings, whose fundamentals all had wavelengths of 11.1° (spatial frequencies, 0.09 cycles/°), and ¾ of the wavelength of the 3f grating, whose wavelength was 3.7° (spatial frequency, 0.27 cycles/°). Traces show the mean R-L eye velocity temporal profiles. The keys indicate the contrasts, those values in parentheses for the mf and 3f4f data indicating the contrasts of the 3^rd harmonic and 3f component, respectively. Upward deflections of the traces from the zero baselines (dashed horizontal lines) denote forward motion (in the direction of the steps). Abscissas denote the time from the first stimulus shift (defined as the onset of stimulus motion).

Let us scrutinize the response profiles in Fig. 2 more closely, starting with those obtained with the 1f stimulus (upper left). At the lowest contrast (1%), the initial transient OFR had a latency of 70–80 ms and reached a peak at ~100 ms after the onset of the apparent motion (i.e., the time of the first step). As the stimulus contrast was increased to 2%, this initial transient showed a slight reduction in latency and a substantial increase in amplitude. Further increases in stimulus contrast resulted in shorter onset latencies with reduced initial peaks, which is the “anomalous contrast dependence” originally described by Miles et al (1986), who recorded the initial OFR elicited from monkeys when velocity steps were applied to pure sine-wave gratings. The response profiles obtained with the 3f stimulus (lower left in Fig. 2) show a similar general pattern even though of the opposite sign, as also do the profiles obtained with the mf stimulus (upper right in Fig. 2) and the 3f4f stimulus (lower right in Fig. 2), consistent with mediation largely by their 3f components. The profiles obtained with the mf stimulus, however, are clearly much lower in amplitude than those obtained with the other stimuli. The OFRs of the other monkey, which are not illustrated, showed the same general tendencies.

The anomalous reductions in the amplitude of the initial peak in the OFR profiles as contrast was increased beyond 2% were attributed by Miles et al (1986) to a reduction in the time available to integrate the motion error signal because of the associated decreases in latency, and they successfully modeled this effect. Indeed, these workers also showed that no such anomalous dependence on contrast was apparent when the velocity responses were integrated over a fixed interval time-locked to motion onset and approximating the initial open-loop period, as though the reductions in velocity amplitude were offset by the reductions in the latency of response onset. We too found this same effect in our data and report it in Fig. 3, which shows the dependence on stimulus contrast of the initial OFR based on the changes in the mean R-L position measures over the time period 50–100 ms (measured from the time of the first step), for all four grating patterns and both monkeys. The R-L response measures for the data obtained with the 1f and 3f stimuli (filled and open circles in Fig. 3), generally showed a monotonic rise as the stimulus contrast increased, gradually saturating as contrast reached 5–10% in monkey A and 20–30% in monkey B (note the log abscissa). Of course, the 1f data all have positive values (being in the forward direction) and the 3f data all have negative values (being in the backward direction). These data were fitted with the following expression:

$$R_{max}\frac{c^n}{c^n+c_{50}^n};$$ (1)

where R_max is the maximum attainable response, c is the contrast, c₅₀ is the semi-saturation contrast (at which the response has half its maximum value), and n is the exponent that sets the steepness of the curves. This expression is based on the Naka-Rushton equation (Naka & Rushton, 1966) and provides a good fit to the contrast dependence curves of neurons in the LGN, V1 and MT of monkeys (Albrecht, Geisler, Frazor & Crane, 2002; Albrecht & Hamilton, 1982; Heuer & Britten, 2002; Sclar, Maunsell & Lennie, 1990), as well as to the human contrast dependence curves for the initial OFRs to moving sine-wave gratings (Sheliga et al., 2005a) and unikinetic plaid patterns (Masson & Castet, 2002). The continuous smooth curves in Fig. 3 are the best fit curves using this expression and are clearly good approximations to the data with r² values >0.9 for all curves. The values of c₅₀ for the 1f and 3f stimuli were 1.47 and 2.01 %, respectively, for monkey A, and 5.02 and 6.16 %, respectively, for monkey B. The values of n for the 1f and 3f stimuli were 3.05 and 1.75, respectively, for monkey A, and 1.14 and 1.98, respectively, for monkey B. Thus, the OFRs were most sensitive to changes in contrast when the contrast was relatively low (<5% for monkey A and <20% for monkey B).

Figure 3.

The initial OFR: dependence on contrast (R-L response measures for two monkeys). Plots show the horizontal OFR elicited when successive steps (each 2.78° for monkey A and 1.67° for monkey B) were applied to mf and 3f4f stimuli (spatial frequencies and wavelengths: monkey A, 0.09 cycles/° and 11.1°; monkey B, 0.15 cycles/° and 6.67°) as well as to pure sine-wave gratings whose spatial frequencies matched the 1f or 3f components of the complex gratings. Responses to the pure 1f sine waves (filled circles) were always positive (OFR in the forward direction), whereas those to the mf stimulus (gray open squares, gray dashed lines), the 3f4f stimulus (gray filled diamonds, gray dotted lines), and the pure 3f sine waves (open circles) gratings were always negative (OFR in the backward direction). Responses to the mf and 3f4f gratings are also replotted as a function of the contrast of their 3f components to permit easy comparison with the pure 3f sine-wave data (mf, black open squares and dashed lines; 3f4f, black filled diamonds and dotted lines). The smooth black curves are best-fit Naka-Rushton functions for the pure sine-wave data and the values of their c₅₀ and n parameters are shown nearby. Monkey A: 40–56 trials per condition; SD’s ranged 0.017–0.062°. Monkey B: 40–54 trials per condition; SD’s ranged 0.038–0.054°. Error bars, 95% confidence intervals.

If the initial OFRs to the mf and 3f4f stimuli are actually generated by their principal Fourier components, the 3^rd harmonic and the 3f component, respectively, then when plotted in terms of the contrast of these components, their contrast-dependence data should overlay those obtained with pure sine waves of the same spatial frequency, i.e., the data obtained with the 3f stimulus. The data obtained with the mf stimulus (squares and dashed lines in Fig. 3) and the 3f4f stimulus (diamonds and dotted lines in Fig. 3) are therefore plotted as a function of both the actual contrast of the patterns (grey symbols and lines) and the contrast of their 3^rd harmonic and 3f components (black symbols and lines). As we pointed out earlier, the OFRs to these mf and 3f4f stimuli were always in the opposite direction to the actual shifts of the patterns and so have negative values in Fig. 3. When plotted as a function of the contrast of their 3f component, the data obtained with the 3f4f stimulus approximated those obtained with the pure sine-wave 3f stimuli, though tending to fall a little short at higher contrasts in monkey B. When plotted as a function of the contrast of their 3^rd harmonic, the data obtained with the mf stimulus also closely approximated those obtained with the 3f stimulus at low contrast (<3% for monkey A, <8% for monkey B), but fell substantially short of this with higher contrast stimuli. Very similar trends are seen in humans (Sheliga et al., 2005a).

For monkey A, the 3f4f data in Fig. 3 closely approximate the pure 3f data (when plotted as a function of the contrast of the 3f component), indicating that the 4f component of the 3f4f stimulus had almost no impact on the initial OFR measures, and this was also apparent from the R-L temporal response profiles: see Fig. 4 (left column) and compare the thick traces (the 3f4f-stimulus data) with the thin traces (the 3f-stimulus data). For monkey B, the 3f4f-stimulus data fell short of the 3f-stimulus data at high contrast and the R-L temporal profiles in Fig. 4 (right column) indicate that the impact of the 4f component was not uniform but rather selectively affected a small early transient component and a later sustained component (that was mostly outside the open-loop measurement period) and left the main body of the response profile largely intact.

Figure 4.

Comparison of the OFRs elicited by 3f and 3f4f stimuli with a range of contrasts (temporal profiles for two monkeys). All shifts and wavelengths were as given in the legend of Fig. 3. Each panel shows the mean R-L temporal velocity profiles elicited by the 3f stimulus alone (thin line), and the 3f4f stimulus (thick line) whose 3f component had a contrast that matched that of the 3f stimulus and is listed on each plot (in %). All responses deflected the traces downward from the zero baseline, denoting backward eye movements (in the direction of the principal Fourier component). Abscissas denote the time from the first stimulus shift (defined as the onset of stimulus motion).

4. Discussion

4.1. The monkey as a model for the human

Our data show that the initial OFRs of the monkey share many fundamental properties with those of the human recently described by Sheliga et al (2005a; 2005b), indicating that the monkey provides an excellent animal model. The initial support for this conclusion comes from the data obtained with the pure sine-wave stimuli and these are further reinforced by the data obtained with the more complex mf and 3f4f stimuli. Let us start by comparing the quantitative R-L response measures obtained with pure sine-wave stimuli in the two studies. The initial OFRs of our monkeys and of Sheliga et al’s humans showed a dependence on spatial frequency that was clearly band-pass and well described by Gaussian functions (with a log abscissa). Even the parameters of the best-fit Gaussians were very similar in the two species. For example, mean values for the 3 humans in the Sheliga et al study vs. mean values for our 2 monkeys were as follows: f_o=0.25 vs. 0.22 cycles/°; s=0.51 vs. 0.51 log units; f_lo=0.06 vs.0.06 cycles/°; f_hi=0.99 vs. 0.87 cycles/°. The dependence on contrast also had a very similar form in monkeys and humans, showing a smooth monotonic rise with saturation at moderate contrast levels that was well fit by the Naka-Rushton equation, whose parameters were again quite similar in our two studies (even though the data from the two monkeys showed much more scatter than the data from the three humans). For example, the mean best-fit Naka-Rushton parameters for the 1f data (human vs. monkey) were as follows: c₅₀ =3.9 vs. 3.24%; n=2.10 vs. 2.09, and the equivalent 3f data are: c₅₀ =5.7 vs. 4.09%; n=1.55 vs. 1.87. Nonetheless, the anomalous dependency on contrast that is evident in the initial peak of the monkey’s eye-velocity traces in both our study and that of Miles et al (1986), whereby over part of the contrast range, increases in contrast were associated with decreases in the amplitude of the initial peak in eye velocity, is not seen in the temporal profiles of the human OFR (Sheliga et al., 2005a). This anomaly presumably had no obvious impact on our integrated-velocity measures that were time-locked to stimulus onset (i.e., the change-in-position measures in Fig. 3) because of the associated latency changes, as originally pointed out by Miles et al (1986). In fact, the latency of the monkey’s OFR (to smoothly drifted sinusoidal gratings) is solely a function of contrast and temporal frequency (Miles et al., 1986), whereas the latency of the human’s OFR is also somewhat sensitive to speed and spatial frequency (Gellman et al., 1990) and less sensitive to contrast (see Fig. 5 of Sheliga et al., 2005a). However, these detailed differences in the latency and development of the very earliest OFRs of monkeys and humans ultimately have little consequence for the overall tracking responses as indicated by the similarity of their integrated velocity measures over the initial open-loop period.

Our study showed that the initial OFRs elicited by the mf and 3f4f stimuli were always in the direction of their principal Fourier components (i.e., the 3f component of the 3f4f stimulus and the 3^rd harmonic of the mf stimulus) rather than in the direction of the overall pattern or feature. In fact, the 4f component of the 3f4f stimulus often had almost no impact on the monkey’s initial OFR, including the temporal profiles (Fig. 4). Thus, when the R-L response measures obtained with the 3f4f stimuli were plotted in terms of the contrast of their 3f component, they matched the measures obtained with the pure 3f stimulus quite closely (except in monkey B at higher contrasts, when they fell slightly short). Similar plots of the data obtained with the mf stimulus—using the contrast of the 3^rd harmonic rather than of the whole pattern—mimicked the plots for the 3f stimuli only at low contrasts and then fell progressively short with higher contrast. These data are all very similar to those obtained on humans by Sheliga et al (2005a), further reinforcing the view that the monkey is a good animal model for the initial OFR of humans.

4.2. The initial OFR: A response to 1^st-order motion energy

Sheliga et al (2005a) argued that the dependence on the motion of the principal Fourier components of the mf and 3f4f stimuli—rather than on the motion of their features—indicated that the motion detectors responsible for the initial OFR in humans do not operate directly on the raw retinal images but rather on a spatially filtered version of those images, as in the well-known 1^st-order energy model of motion detection that relies on oriented spatio-temporal filters: see Lu and Sperling (2001) for review. Using mf stimuli lacking the 5^th and 7^th harmonics, Sheliga et al (2005a) were also able to show that the initial OFRs elicited by the usual mf stimulus fell short of those elicited by the pure 3f stimulus mostly because of the higher harmonics, with perhaps a very minor contribution from distortion products due to an early compressive nonlinearity in the visual pathway. Thus, Sheliga et al largely ruled out even a minor rôle for feature-based mechanisms in the genesis of the initial OFR of humans. Although we do not have data for the monkey that clarifies the role of the higher harmonics, we have demonstrated that the monkey’s initial OFR shares much in common with that of the human. Further, the similarity of the contrast dependence of the 3f4f and pure 3f data is consistent with mediation by oriented spatio-temporal filters as in the 1^st-order motion-energy model and indicates that feature-based mechanisms make at best only a minor contribution to the monkey’s OFR with this 3f4f stimulus.

Sensitivity to low-contrast stimuli (<20%), such as we here report for the monkey’s OFR, is considered one of the characteristics that, in humans, sets the 1^st-order motion energy mechanisms apart from feature-based mechanisms (Lu & Sperling, 1995; Nishida, 1993; Smith, 1994; Solomon & Sperling, 1994; Takeuchi & De Valois, 1997). Consistent with this, the study of O’Keefe and Movshon (1998) on monkeys showed that MT neurons have poor contrast-sensitivity for 2^nd-order motion. Yet, Benson and Guo (1999) reported that the initial OFRs to a pure 2^nd-order motion stimulus (defined by contrast modulated noise) were little different from those to a stimulus with strong 1^st-order motion energy (except for a small latency difference, averaging ~11 ms). This observation, which was made only on a single monkey, would seem to be at odds with our findings, perhaps suggesting that some types of 2^nd-order motion are much more effective in initiating OFR than others.

5. Closing Remarks

Earlier studies in monkeys (Busettini, Miles & Schwarz, 1991; Kawano & Miles, 1986; Miles et al., 1986) and humans (Busettini, Miles, Schwarz & Carl, 1994; Gellman et al., 1990) demonstrated a number of functional similarities between the OFRs of the two species which fostered the idea that the monkey was a good animal model for the human. Our present findings on the initial OFRs elicited in monkeys by motion applied to complex grating patterns are largely in accord with the findings in a recent study on humans that used these same visual stimuli (Sheliga et al., 2005a), suggesting that the similarities between the two species extend to the detectors whereby they sense visual motion.