We characterize a hitherto undocumented type of neuron present in the regions bordering the principal layers of the macaque lateral geniculate nucleus. Neurons of this type were distinguished by a high and unusually regular maintained discharge that was suppressed by spatiotemporal modulation of luminance or chromaticity within the receptive field. The response to any effective stimulus was a reduction in discharge, reminiscent of the “suppressed-by-contrast” cells of the cat retina. To a counterphase-modulated grating, the response was a phase-insensitive suppression modulated at twice the stimulus frequency, implying a receptive field comprised of multiple mechanisms that generate rectifying responses. This distinctive nonlinearity makes the neurons well suited to computing a measure of contrast energy; such a signal might be important in regulating sensitivity early in visual cortex.
The principal projections from the primate retina to the dorsal lateral geniculate nucleus (LGN) originate in the midget and parasol ganglion cells, and terminate in the parvocellular (P) and magnocellular (M) cells, respectively. Together, these streams comprise ∼90% of all of the relay neurons found in the LGN (Perry et al., 1984; Lee, 1996; Solomon, 2002; Wassle, 2004). P and M cells have center-surround receptive fields in which spatial summation is generally well described by linear models (Kaplan et al., 1990; Lee, 1996).
Midget and parasol ganglion cells are two of the ∼15–20 morphological classes of ganglion cells in the primate retina. Other classes also project to the LGN (Dacey et al., 2003; Wassle, 2004), among them neurons that draw substantial input from short-wavelength-sensitive (S) cones, and whose signal summation within receptive fields is also generally well described by linear models (Derrington et al., 1984; Dacey and Lee, 1994; Martin et al., 1997; Dacey et al., 2003). Neurons with patently nonlinear receptive fields also exist (Schiller and Malpeli, 1977; Casagrande, 1994; White et al., 2001), but they have not been characterized, and are not accommodated in models of visual perception.
Here, we characterize a distinctive and homogeneous class of neuron in the macaque LGN. This has a regular and rapid spontaneous discharge that is suppressed by spatiotemporal contrast presented at any location within the receptive field. They resemble the “suppressed-by-contrast” neurons found in the cat retina and LGN (Rodieck, 1967; Stone and Hoffmann, 1972; Cleland and Levick, 1974).
Materials and Methods
Extracellular single-unit recordings were made from LGNs of eight anesthetized, paralyzed juvenile male Macaca fasicularis. The preparation and recording were as described previously (Solomon and Lennie, 2005). The microelectrodes were tungsten-in-glass (Merrill and Ainsworth, 1972). All procedures conformed to the guidelines approved by the New York University Animal Welfare Committee. Stimuli were presented on a calibrated cathode ray tube monitor (Vision Master Pro 514; Iiyama, Nagano, Japan) refreshed at 200 Hz and viewed from 114 cm, at which distance the screen subtended 15° × 20°. For some neurons, we also measured responses at lower refresh rates. Stimuli were defined by spatiotemporal modulations of luminance or chromaticity about a white point (Commission Internationale de l'Eclairage coordinates, x = ∼0.30, y = 0.32) at a mean luminance of ∼33 cd/m−2. The stimulus was an 8°-diameter patch containing a drifting sinusoidal grating or a temporally modulated spatially uniform field. The rest of the screen was held at the mean luminance. Unless otherwise stated, the stimulus was achromatic and of full contrast. From the impulses discharged by each cell, we extracted the mean rate and the amplitude of the Fourier component at the frequency of stimulation (and in some cases the amplitude at twice that frequency). To characterize the tuning curves, we fit models of the form Rpred = ⌊MD − F(x)⌋+, where Rpred is the predicted mean discharge in the presence of the stimulus, MD is the maintained discharge to the mean luminance, F(x) is a function describing tuning along the stimulus dimension of interest, and the operation ⌊ ⌋+ sets negative responses to 0 (rectification). We derived model fits by minimizing the χ2 error between the model predictions and the observed discharge rate in Matlab (version 7.0; Mathworks, Natick, MA).
The 12 neurons whose receptive fields are described here were encountered during a larger series of experiments on LGN. The first five neurons were fortuitously discovered in one monkey, and in subsequent experiments on seven animals we directed electrode penetrations to maximize the chance of encounter. We found seven more neurons in four of the monkeys. Throughout we will refer to these neurons as “suppressed-by-contrast” cells, because in many respects their receptive fields resemble those of the cat retinal ganglion cells described by Rodieck (1967).
The maintained discharge of suppressed-by-contrast cells was distinctively high, averaging 45.4 spikes/s (SD, 12.7; n = 12) when the animal viewed a uniform gray screen refreshed at 200 Hz. Under the same viewing conditions, the average maintained discharge of P-cells was 15.9 spikes/s (SD, 10.9; n = 250), of blue-yellow opponent cells, 20.0 spikes/s (SD, 12.2; n = 87), and of M-cells, 15.1 spikes/s (SD, 9.9; n = 113) (Troy and Lee, 1994). The interspike interval histograms and autocorrelograms in Figure 1A–C make clear that the maintained discharge of suppressed-by-contrast cells is also much more regular than that of P and M cells measured under the same conditions. This regularity can be characterized by the coefficient of variation (CV: the SD of the interval distribution divided by its mean; for a random spike train, the CV is 1 and for a regular spike train the CV is 0). In suppressed-by-contrast cells, the average CV was 0.61 (SD, 0.23), significantly lower than the corresponding CV values for P cells (μ = 0.87; SD, 0.27; n = 193), M cells (μ = 0.86; SD, 0.17; n = 64), and blue-yellow opponent cells (μ = 0.95; SD, 0.40; n = 80) (Troy and Lee, 1994). The regularity of the discharge of suppressed-by-contrast cells was unrelated to the monitor refresh rate (200 Hz): the modal interspike interval for the cell shown in Figure 1C is 16.6 ms; across all cells, the modal interval ranged from 13.6 to 22.9 ms (median, 15.4; SD, 2.8).
In every suppressed-by-contrast cell, the spontaneous discharge was reduced by the presence of spatiotemporal contrast within the receptive field, regardless of the polarity of that contrast. Figure 1F shows for one cell the pattern of discharge during square-wave luminance modulation of a large disc centered on its receptive field. Transient reductions in discharge occurred at both the on and off phases of the modulation. This behavior is fundamentally different from that of M and P cells: the discharge of an on-center P or M cell (Fig. 1E) increases in response to a luminance increment and decreases in response to a luminance decrement (vice versa for an off-center cell) (Fig. 1D). We found no stimulus that increased the discharge of a suppressed-by-contrast cell above the maintained rate (see Discussion).
All suppressed-by-contrast cells had monocular receptive fields: three had receptive fields in the contralateral eye and were encountered in the region separating the two most ventral P layers, and the remaining nine had receptive fields in the ipsilateral eye and were encountered in the region separating the most ventral P layer and the dorsal M layer. We did not make lesions at the recording sites to preserve the electrodes, but the locations were unambiguous given the pattern of transitions between eye representations, the low background activity at the transitions, and the functional properties of the cells on either side of these transitions. The receptive fields of the suppressed-by-contrast cells were located at eccentricities between 2 and 15° in the contralateral hemifield in positions consistent with those of M and P cells found near them.
Spatial frequency tuning
The spatial frequency tuning curves of suppressed-by-contrast cells were weakly bandpass and very similar from cell to cell (Fig. 2A, individual cell, B, average across the 10 cells for which we obtained these measurements). At mid to high spatial frequencies, the mean rate of discharge was suppressed uniformly throughout the drift cycle; there was no modulated component of the discharge of the kind that characterizes P and M cells. Very low spatial frequencies (including modulation of a spatially uniform field) reduced the mean discharge while also modulating the discharge at the stimulus frequency. The modulated response to a uniform field resembled half-wave rectification of the input signal: discharge was suppressed completely during the off phase of the modulation and less, if at all, during the on phase. At no point in the cycle did the discharge exceed the spontaneous rate (Fig. 2A, PSTH, inset). No suppressed-by-contrast cells were orientation selective. In what follows, we use the reduction in mean discharge rate to measure the stimulus selectivity of the suppressive mechanism.
To characterize the spatial frequency tuning, we fit the suppression with a difference of Gaussians model (D-o-G) (Enroth-Cugell and Robson, 1966). The most suppressive spatial frequency was 1.9 cycles/deg (geometric mean, n = 10), similar to the optimum spatial frequency for modulation of P cells and exceeding that of M cells. These relationships are reflected in the center sizes returned by the D-o-G model: for suppressed-by-contrast cells, the geometric mean center radius was 0.064° (SD, 0.03), and for P cells and M cells at comparable eccentricities, 0.067° (SD, 0.05; n = 170) and 0.094 deg (SD, 0.04; n = 61).
Temporal frequency tuning
Suppressed-by-contrast cells had bandpass temporal frequency tuning curves (Fig. 2C, individual cell, D, average of six cells). To characterize temporal tuning, we used a descriptive function (Webb et al., 2005). Across all cells, suppression was strongest at 14.6 Hz (geometric mean, n = 6). No cell was suppressed by counterphase modulation at 100 Hz (data not shown). Gratings drifting at low temporal frequency (0.5 Hz) were less effective, with the reduction in discharge being 69% of the maximum suppression observed. The temporal tuning lay between that of P and M cells: P cells preferred 12.4 Hz, and the response at the lowest frequency was 46% as effective as at the best (n = 251); M cells preferred 25.0 Hz, and the response at the lowest frequency was 25% as effective as at the best (n = 110) (Derrington and Lennie, 1984).
We measured the latency of suppression after presentation of a stimulus, and the latency of the increase in discharge after removal of a stimulus (Muller et al., 2003). The mean latency of suppression was 39.0 ms (SD, 3.8; n = 11) and the mean latency of increased discharge was 51.2 ms (SD, 12.9). For comparison, the latency of a comparably strong increment step response in on-center P cells was 40.5 ms (SD, 8.8; n = 27), and in on-center M cells was 31.9 ms (SD, 6.4; n = 21).
Figures 2, E and F, shows how the mean discharge rate declined with increasing contrast of drifting gratings of the most effective spatial frequency (Fig. 2E, individual cell, F, average of eight cells). Over most of the range, the mean rate was inversely related to contrast. We fit a Naka-Rushton function to the data to determine the average contrast gain (the slope of the initial part of the contrast response function), which was −1.8 impulses/s/[% contrast]. This is higher than the average contrast gain of P cells (1.1), but substantially lower than that of M cells (5.9) (Kaplan and Shapley, 1986; Croner and Kaplan, 1995; Levitt et al., 2001).
Drifting gratings suppressed the discharge of a suppressed-by-contrast cell but did not usually modulate it, suggesting that the underlying mechanism is insensitive to the spatial phase of the stimulus. We confirmed this in seven cells by measuring discharge to counterphase-modulated gratings of the most suppressive spatial and temporal frequencies. Figure 3A shows the results for one cell. At all spatial phases, the mean rate of discharge was reduced, with little modulation at the frequency of stimulation (F1) and substantial modulation at twice that frequency (F2). There was no “null” phase at which the discharge was unaffected by the stimulus. We used an index devised by Hochstein and Shapley (1976a) (the average of the F2 across spatial phases divided by the maximum F1 at any phase) to quantify the nonlinearity of spatial summation. The geometric mean index was 1.7 (n = 7), much greater than that for P or M cells (average 0.42 and 0.63, respectively) (Levitt et al., 2001). The phase-insensitive rectified responses to counterphase-modulated gratings, coupled with the high spatial resolution (Fig. 2), implies that the receptive field of the suppressed-by-contrast cell is comprised of an array of relatively small rectifying spatial subunits (Hochstein and Shapley, 1976b; Troy et al., 1989).
Receptive field size
The size of a linear receptive field characterized by the D-o-G model can be estimated from the spatial frequency tuning curve. To estimate the size of the nonlinear receptive field of a suppressed-by-contrast cell, we obtained a size-tuning curve using patches of drifting grating of the most suppressive spatial and temporal frequencies. The set of responses was fit with the integral of a two-dimensional Gaussian (Fig. 3B), the characteristic radius of which was used as the measure of receptive field size. The geometric mean radius was 0.143° (SD, 0.083; n = 7). The geometric mean subunit size, inferred from spatial frequency measurements in the same cells was 0.055° (SD, 0.024). These measurements of receptive field size may be underestimates, because they were made with high-contrast stimuli, and the discharge was sometimes driven to zero for larger stimulus diameters.
The receptive fields of some retinal ganglion cells have distinctive chromatic signatures: P (midget) cells compare the signals of long-wavelength-sensitive (L) and middle-wavelength-sensitive (M) cones; the small-bistratified cell receives strong excitatory input from S cones (Dacey and Lee, 1994). To characterize the chromatic signatures of suppressed-by-contrast cells, we measured responses to gratings modulated in different color directions about a white point (Derrington et al., 1984). In all 10 cells on which measurements were made, responses were suppressed strongly by isoluminant L–M modulation and by achromatic modulation, but only weakly by S-cone modulation. The signs and weights of cone inputs were estimated from a model that assumes a weighted sum of the signals of L, M, and S cones is full-wave rectified and subtracted from the maintained discharge (Lennie et al., 1990). We also explored a model in which cone signals were rectified before rather than after summation; this proved inadequate. The normalized weights of L- and M-cone inputs to the receptive fields of suppressed-by-contrast cells were on average 0.37 and −0.58, respectively (the signs are opposite, with that of the L cone arbitrarily positive), and for S cones the unsigned weight was 0.03. Suppressed-by-contrast cells are therefore substantially driven by opposed signals from L and M cones.
We have characterized a distinctive and homogeneous class of cell not encountered previously in the LGN of macaque. We saw S potentials associated with the spikes recorded from three of our neurons and, although these were not well enough isolated to characterize the ganglion cell receptive fields, we think it likely that the distinctive receptive field properties are determined in the retina; De Monasterio and Gouras (1975), Schiller and Malpeli (1977), and de Monasterio (1978) have described primate ganglion cells whose discharge appeared to be suppressed by any visual stimulus.
The receptive fields of the cells we encountered resemble those of some rarely encountered retinal ganglion cells in cat [the suppressed-by-contrast cells (Rodieck, 1967; Stone and Hoffmann, 1972; Troy et al., 1989; Rowe and Cox, 1993); the uniformity detectors (Cleland and Levick, 1974)] and in rabbit (Levick, 1967; Amthor et al., 1989). Stone and Hoffmann (1972) observed that after removal of spatiotemporal contrast, the discharge of a suppressed-by-contrast cell was often briefly elevated above the maintained rate. We also saw this, and found that extending the duration of exposure to the suppressive stimulus led to greater poststimulus discharge, suggesting that it reflects habituation of the suppressive mechanism.
A complementary pathway that is excited by the presence of spatiotemporal contrast and insensitive to spatial phase (the “impressed-by-contrast” or “on–off” retinal ganglion cells) is present in the retinas of cat (Cleland and Levick, 1974; Troy et al., 1989) and monkey (De Monasterio and Gouras, 1975; Schiller and Malpeli, 1977; White et al., 2001). We encountered two apparently similar neurons in the macaque LGNs, but have not studied them systematically.
Receptive field model
Troy et al. (1989) provided a receptive field model for suppressed-by-contrast cells in cat, based on an earlier model of the Y cell developed by Hochstein and Shapley (1976b). In this model, the rectified activity of an array of subunits is subtracted from a maintained discharge. The subunits of the model presumably reflect bipolar cells whose normally excitatory output is rectified, as in Y cells (Demb et al., 2001), and inverted, either through an amacrine cell or directly via an inverting synapse. This model offers a potentially promising account of the observations here: the subunits provide fine spatial resolution and their rectified outputs make the suppressed-by-contrast cell sensitive to spatial frequencies beyond that expected from the size of its receptive field. If the subunits have a center-surround receptive field organization and their centers are small enough, drawing input from only a few cones, they would be sensitive to chromatic modulation (as has been proposed for P cells) (Lennie et al., 1991). The temporary increase in discharge after a prolonged stimulus might reflect a slowly recovering hyperpolarization in the bipolar cells themselves (Manookin and Demb, 2006).
Temporal modulation of a uniform field produced responses that do not fit easily with this model. We saw suppression at both on and off phases of square wave modulation (Fig. 1C), but for sinusoidal modulation there was no temporal frequency that gave rise to modulation at twice the frequency. At low temporal frequencies, the discharge rate dropped only during the off-phase of stimulation (Fig. 2A) (Rowe and Cox, 1993); at higher frequencies, the discharge was much lower than the maintained rate throughout the cycle, and the modulation disappeared at rates of > 5 Hz (data not shown). The model above can produce the modulation at low temporal frequencies if the subunits are all off-center and their outputs are half-wave rectified (Demb et al., 2001) before subtraction from the maintained discharge, but this does not explain the responses at higher temporal frequencies, or the on–off suppression during square-wave modulation (Fig. 1). A small extension to the model, captured in Figure 4, might accommodate our results. We think that suppression arises in the rectified responses of a collection of on- and off-center subunits, all of which show strong spatial opponency and are thus largely ineffective at low spatial frequencies (the transient suppression for square wave modulation arising because the surround of the subunit slightly lags the center). Their summed output is subtracted from the maintained discharge, which is itself driven by a light-dependent excitatory mechanism that saturates. The discharge during sinusoidal modulation of a uniform field reflects the properties of this mechanism: the saturation prevents the discharge during modulation from ever exceeding the maintained rate and, at higher temporal frequencies, its effect is equivalent to a reduction in mean luminance (the mechanism is more sensitive to decrements than increments).
Why have suppressed-by-contrast cells not been seen before in primate LGN? One reason may be that low monitor-refresh rates suppress the maintained discharge, from ∼45 spikes/s at 200 Hz to ∼20 spikes/s at 60 Hz (because the neurons respond to the frame updates). This removes the distinctiveness of the maintained discharge. Moreover, the usual practice of presenting contrast stimuli when searching for neurons while advancing the microelectrode will tend to silence a suppressed-by-contrast cell: brief decreases in discharge are harder to detect than brief increases. Finally, the location of these cells in the interlaminar regions suggests that they have small cell bodies that are consequently hard to record from (Rodieck, 1967; Cleland and Levick, 1974). Were the retinal coverage of suppressed-by-contrast cells the same as that of M cells, the relative densities of the two cell types could be estimated from their receptive field areas. Using our measurements of receptive field size of suppressed-by-contrast cells, and those of Croner and Kaplan (1995) for M cells, suppressed-by-contrast cells might be half the density of on-center (or off-center) M cells, and therefore probably no more than 2% of all ganglion cells.
The retinogeniculate pathway is the main pathway by which information is transmitted from retina to visual cortex, and therefore the pathway most associated with visual awareness. The presence of suppressed-by-contrast cells in the LGN suggests that they might have some relatively direct access to visual sensation, although what their role might be is unclear. We would be better placed to understand this if we knew how suppressed-by-contrast cells behaved in natural visual environments: are they normally suppressed by the pattern of spatiotemporal contrasts imposed on their receptive fields by normal eye-movements, so that the absence of contrast (perhaps associated with a saccade) generates a discharge?
Cleland and Levick (1974) suggested a spike discharged by a suppressed-by-contrast cell indicates “no recent spatial or temporal discontinuities.” Apart from the sign of its response, the suppressed-by-contrast cell differs from other known LGN cells in two notable ways: the regularity of its discharge and its indifference to stimulus phase. The regularity of discharge makes changes in firing rate potentially more detectable than would be the case if the maintained discharge resulted from a renewal process with the same mean rate and variability (Troy et al., 1989). The indifference to stimulus phase suggests that the cell can be thought of as computing a measure of contrast energy. Such a mechanism could be useful in setting contrast gain (Troy et al., 1989). This is an appealing possibility because it explains why it might be useful to have a highly nonlinear signal available as input to cortex.
This work was supported by National Institutes of Health Grants EY04440 and EY13079, and Australian National Health and Medical Research Council Grant 211247 to S.G.S. We thank J.A. Movshon and W. Levick for helpful discussions.
- Correspondence should be addressed to Chris Tailby at the above address.