Differential attentional modulation of cortical responses to S-cone and luminance stimuli

Abstract

Neural signals driven by short-wave-sensitive (S) cones are, to a large degree, anatomically and functionally separate from the achromatic luminance pathway until at least one synapse into V1. Attentional mechanisms that act at an anatomically early stage in V1 may, therefore, affect S-cone and luminance signals differently. Here, we used a steady-state visually evoked potential (SSVEP) paradigm combined with electrical source imaging to study the effects of contrast and attention on neural responses to both chromatic S-cone isolating and achromatic stimuli in five human visual areas including V1. The responses to these gratings were affected very differently by changes in contrast and attention. Increasing cone contrast increased the response amplitude for both types of stimulus. For the S-cone-defined stimuli, we also observed a systematic decrease in the response phase of the first harmonic with increasing stimulus contrast, but there was no corresponding change in phase for the first harmonic of the luminance probes. Attending to the contrast of the grating increased the amplitude and phase of luminance-driven responses but had no effect on S-cone-driven responses. We conclude that while attentional modulation can be observed in achromatic pathways as early as V1, attention may not affect SSVEP signals generated by S-cone stimuli.

Introduction

Recent computational theories of spatial and featural attention in the visual system propose that attentional modulation acts to drive a multiplicative “gain field” prior to a cortical contrast normalization stage (Boynton, 2009; Reynolds & Heeger, 2009). The natural site for an early gain field of this type is V1 and there is recent fMRI evidence that the striate cortex may be modulated by attention in a manner consistent with this theory (Herrmann, Montaser-Kouhsari, Carrasco, & Heeger, 2010). 

However, single unit studies in macaque disagree as to the presence and amount of attentional modulation in V1. Some report statistically significant effects (Chen et al., 2008; Herrero et al., 2008; Ito & Gilbert, 1999; McAdams & Reid, 2005; Motter, 1993; Roelfsema, Lamme, & Spekreijse, 1998; Thiele, Pooresmaeili, Delicato, Herrero, & Roelfsema, 2009), while others find little evidence of early attentional gain control (Luck, Chelazzi, Hillyard, & Desimone, 1997; Marcus & Essen, 2002; McAdams & Maunsell, 1999; Moran & Desimone, 1985). Human neuroimaging experiments are less equivocal and invariably find robust attentionally driven changes in the primary visual cortex (Buracas & Boynton, 2007; Herrmann et al., 2010; Li, Lu, Tjan, Dosher, & Chu, 2008; Tootell et al., 1998) and even in subcortical structures (O'Connor, Fukui, Pinsk, & Kastner, 2002; Schneider & Kastner, 2009). Finally, human electrophysiological studies often report (Clark & Hillyard, 1996; Luck & Hillyard, 1994; Mangun, Buonocore, Girelli, & Jha, 1998; Martínez et al., 1999; Russo, Martínez, & Hillyard, 2003; Wang, Clementz, & Keil, 2007) attentional modulation in the visual cortex, but the spatial origins of these signals can usually only be estimated by indirect means (for example, by inspection of signal timing or the dependence of waveforms on stimulus location). Direct estimates of neural modulation from electrophysiological measurements with higher spatial precision differ considerably. For example, our group has recently demonstrated robust attentional modulation in V1 using a source imaging method similar to the one described in this paper (Lauritzen, Ales, & Wade, 2010). 

However, another recent study of attentional modulation in human V1 using intracranial electrodes in a single subject failed to find an effect of attention (Yoshor, Ghose, Bosking, Sun, & Maunsell, 2007). One possible explanation for this finding is that the attentional task was not demanding enough to generate strong neural response modulation. In general, it appears that attention can act at early sites in the visual stream and modulate neural responses to achromatic contrast, but the effects may be weaker than those seen in higher areas or confined to a subset of the neural population. 

With some important exceptions (Di Russo & Spinelli, 1999b; Di Russo, Spinelli, & Morrone, 2001; Highsmith & Crognale, 2010), most studies of attentional modulation of neural response in humans have used achromatic stimuli. The effect of attention on chromatic signals, particularly those carried by the koniocellular pathway, is less well studied. This issue is interesting because it may help identify the anatomical site of attentional modulation in V1. 

Signals from the three distinct visual pathways present in the LGN enter V1 in different layers. The majority of neurons in the magnocellular and parvocellular pathways target neighboring laminae of layer 4C (4Cα and 4Cβ, respectively). However, the majority of koniocellular pathway axons, carrying signals initiated in the retinal short-wave-sensitive “S” cones, appear to enter the striate cortex in layer 2/3 (Hendry & Reid, 2000; Sincich & Horton, 2005). 

Because of the KC pathway's unique anatomical properties, it is possible that attention acts on it very differently. Signals from the KC, MC, and PC pathways are combined no later than two synapses into V1 and so an attentional effect that modulates the latter but not the former must act at the earliest stages of cortical processing. In effect, the koniocellular pathway provides a way to study and perhaps isolate the earliest sites of visual attention. 

Previous research using scalp EEGs has shown that signals driven by pure achromatic luminance and pure chromatic (L–M)-cone signals are modulated in different ways by attention. Specifically, Di Russo and Spinelli (1999b) and Di Russo et al. (2001) examined the effects of spatial attention on steady-state contrast-reversing gratings defined by either luminance or isoluminant (L–M)-cone contrast. They found that both types of signal experienced response gain control in the presence of attention resulting in an amplification primarily of the high, but not the low, end of the contrast response function. Both pathways experienced qualitatively similar amplitude modulations, but only the luminance signals also experienced a change in response phase due to attention. Although the stimuli used in these studies could not isolate the parvocellular and magnocellular pathways perfectly, the authors hypothesized that attention affected these pathways in different ways. Paradoxically, another recent study found no effect of attention on either the amplitude or phase of L–M or S-cone isolating chromatic responses (Highsmith & Crognale, 2010). Importantly, neither study was able to identify the cortical source of the visually evoked signals. 

In this paper, we defined a set of early visual areas using standard fMRI retinotopic mapping techniques. We then used electrical source imaging (ESI) combined with steady-state visually evoked potentials (SSVEPs) to compare the effects of attention on neural signals generated by achromatic and S-cone isolating stimuli in multiple visual areas. We measured these effects at different contrast levels to measure the way in which attention alters the contrast-versus-response curves of neural populations in different areas and fit these functions using hyperbolic ratio functions (Albrecht & Hamilton, 1982; Naka & Rushton, 1966) to parameterize the type and size of the effects. 

Methods

Participants

Twelve participants (3 females, mean age = 34 years) participated in this study. All participants had normal or corrected-to-normal vision (visual acuity of 20/20). The Smith-Kettlewell Eye Research Institute Review Board approved this study; participants provided informed consent before testing. Data from one subject were rejected after acquisition because of poor signal quality. 

Stimuli and procedure

Our visual stimuli consisted of an annulus (width: 4°; inner radius: 1°) containing a two cycle per degree vertical grating defined by either achromatic luminance contrast or chromatic S-cone isolating contrast (Figure 1). The grating was flickered on and off at 6 Hz over the course of each 12-s trial presentation. Participants were required to fixate on a central point at all times. A stream of letters was presented around this point and entirely contained within the center of the annulus. We modulated attention throughout our experiments by asking subjects to perform one of two potential tasks on each trial. 

Figure 1

 

Example of stimulus. A grating masked by a parafoveal annulus flickered on and off at 6 Hz. Before the trial started, subjects were instructed to attend either to the contrast of the grating or to detect “Ts” in a field of “Ls.” Stimuli could be defined by either S-cone contrast (shown here) or achromatic contrast that stimulated all cone classes.

In the “Contrast detection” task, participants were instructed to respond with a button press when they detected near-threshold contrast decrements. Contrast decrement amplitudes were determined outside the EEG system on a subject-by-subject basis to achieve a mean 78% correct detection rate. 

In the “Letter detection” task, participants were instructed to attend to letter stream and detect a randomly oriented probe letter “T” among a field of “L” shapes (see Figure 1). Letter arrays containing “Ls” and “Ts” were preceded and followed by masking arrays of only “Fs.” Participants pressed one button when detecting target letter “T,” otherwise pressed another button. A staircase procedure was used to alter the stimulus presentation duration (typically around 100 ms) to maintain a constant, high level of task difficulty (75% correct). 

Responses to both the achromatic and chromatic stimuli were recorded at four contrast levels (luminance: 5%, 10%, 20%, 40%; S-cone: 10%, 20%, 40%, 80%). Each contrast level was repeated 10 times and time course data from repeated measurements were averaged coherently. 

Cone isolation

Our stimuli and surrounds were defined by contrast modulations in one of two directions in cone contrast space. One direction was designed to stimulate only the S-cones, leaving the quantal catch in the L- and M-cones unaffected, thereby generating a strong and independent signal in the opponent S–(L + M) pathway. The other direction was designed to stimulate all cone classes equally, nulling the chromatic pathways and generating a signal in the achromatic luminance pathway. Subjectively, these stimuli appeared to modulate between lilac and chartreuse or between black and white, respectively, when viewed on the mean gray background. 

Stimuli were displayed on a Lacie Electron Blue II CRT monitor at a frame rate of 72 Hz and a resolution of 800 × 600. 

Cone isolation was achieved using a silent substitution technique (Estevez & Spekreijse, 1982) in which cone contrasts were converted to monitor RGB values via a transform matrix computed from the monitor spectra measured using a spectroradiometer (USB2000, Oceanoptics, Dumolin FL) and the Stockman cone fundamentals (Stockman, MacLeod, & Johnson, 1993) appropriate for the eccentricity of our stimulus. Subject-to-subject correction for perceptual isoluminance was performed using an adjustable minimum motion psychophysical paradigm (Anstis & Cavanagh, 1983). In this procedure, subjects adjusted the amount of achromatic luminance contrast in a drifting S-cone grating dynamically until the grating speed was minimized. This procedure was repeated 5 times and the average cone contrast of all match points was used as the S-cone isolating axis. All subjects made consistent matches very close to the nominal S-cone isolating direction and reported a percept of both strong motion nulling and reduced grating visibility at the isoluminant point. The average CIE (x, y) coordinates of the S-cone isolating axis defined in this manner were +S (0.275, 0.272) and −S (0.301, 0.374) on a mean gray background with coordinates of (0.286, 0.312) and luminance of 57.1 cd/m². The letters used in the attentional task had a chromaticity of (0.285, 0.312) and a luminance of 113.8 cd/m². 

Our 8-bit display hardware and our gamma correction imposed a limit of 1.5% maximum error in our RMS cone contrast specification leading to small luminance artifacts in our nominally isoluminant stimuli. In addition, although our stimuli were chosen to reduce the possibility of chromatic aberration—extending to only 5° in radius and containing a relatively low spatial frequency grating—it is possible that optical effects contributed to small deviations from isoluminance. 

These small luminance artifacts are unlikely to contaminate our data for two reasons. 

First, we have verified that we do not measure significant responses to true luminance contrast at these low levels (up to 2% achromatic cone contrast). In particular, the first harmonic responses to achromatic luminance contrast are relatively weak compared to the S-cone responses (see Figure 3)—presumably reflecting the differences between the degree of sustained and transient responses in the different pathways (Kulikowski, McKeefry, & Robson, 1997; Kulikowski, Robson, & McKeefry, 1996). In general, S-cone-driven SSVEP signals have been shown previously to be relatively robust to small amounts of luminance contamination (Rabin, Switkes, Crognale, Schneck, & Adams, 1994). 

Second, we measure qualitatively and quantitatively different responses to S-cone and achromatic stimuli suggesting that our S-cone data are not simply reflecting a weak contribution from the achromatic luminance pathways. 

EEG recording

Scalp EEG data were measured using a 128-channel Geodesic Sensor Net and NetAmps 200 amplifiers (Electrical Geodesics, Inc. (EGI)) at a sampling rate of 432 Hz. For the purposes of early signal processing and artifact rejection, the recordings were referenced to the vertex sensor (Cz) although in our final source imaging procedure the precise choice of inverse is unimportant. As is standard with high input impedance amplifiers like those from EGI, sensor impedances were <60 kΩ. Data were recorded continuously throughout the experiment. They were analog filtered from 0.1 to 100 Hz and stored on disk for later offline analysis. Following each experimental session, the 3D locations of all electrodes and three major fiducials (nasion, left, and right peri-auricular points) were digitized using a 3Space Fastrack 3D digitizer (Polhemus, Colchester, VT). For all participants, the 3D digitized locations were used to coregister the electrodes to their T1-weighted anatomical magnetic resonance (MR) scans. 

EEG data analysis

Raw data were checked for bad channels (<5% of the total for any participant). Bad channels were automatically replaced by the average of the six nearest spatial neighbors. Data were transformed to an average reference and digitally high pass filtered at 1 Hz (12 dB/octave roll-off, zero phase) combined with 50 Hz (±2 Hz) low pass filter. After preprocessing, data were binned in 1-s bins according to condition (contrast, color/achromatic, and attentional state). A total of 12 bins were acquired for each presentation of each condition and the first and last bins were discarded to eliminate transients. Time series data were averaged within the remaining ten 1-s bins and a discrete Fourier transform was used to generate spectral responses at a resolution of 1 Hz. We analyzed data at and around the first harmonic (6 Hz) and second harmonic (12 Hz) with side bins (5, 7, 11, 13 Hz) providing an estimate of local noise. One of our 12 subjects exhibited almost no response (SNR < 1) at either the first or second harmonic and was removed from subsequent analysis. 

Head conductivity and geometry models

As part of the source estimation procedure, head tissue conductivity models were derived for each individual from T1- and T2-weighted MR scans at a resolution of 1 × 1 × 1 mm. Boundary element models were computed based on compartmentalized tissue segmentations that defined contiguous regions for the scalp, outer skull, inner skull, and the cortex. To begin, approximate cortical tissue volumes for gray and white matter were defined by voxel intensity thresholding and anisotropic smoothing using the FSL package (http://fmrib.ox.ac.uk). The resulting white matter tissue boundaries were used to extract the contiguous cortical gray matter surface. These approximate segmentations were then used as a starting point for the anatomical segmentation procedure in Freesurfer (http://surfer.nmr.mgh.harvard.edu). This package used an iterative mesh-fitting procedure to generate topologically correct estimates of the white matter surface, the pial surface, as well as the inner and outer skull boundaries and the scalp. We used a surface midway between the white matter and the pial surface as our boundary element model cortex. 

Finally, all tissue surface tessellations were visually checked for accuracy to assure that no intersection had occurred between concentric meshes. Coregistration of the electrode positions to the MRI head surface was achieved by aligning the three digitized fiducial points with their locations on the anatomical MR head surface using a least-squares algorithm in Matlab. Minor electrode deviations from the scalp surface were removed by projecting each electrode onto the scalp mesh along the surface normal. 

Cortically constrained minimum norm source estimates

Estimates of the underlying cortical activity based on measurements of the potential recorded at the scalp were derived using the cortically constrained minimum norm estimate (Hämäläinen & Ilmoniemi, 1994) of the MNE package (http://www.nmr.mgh.harvard.edu/martinos/userInfo/data/sofMNE.php). This technique assumes that surface EEG signals are generated by multiple dipolar sources that are located in the gray matter and oriented perpendicular to the cortical surface. Cortical Current Density (CCD) estimates were determined based on an iterative approach that attempts to produce a continuous map of current density on the cortical surface having the least total (RMS) power while still being consistent with the voltage distribution on the scalp (Figure 2). 

Figure 2

 

Visual areas used to define ROIs for cortical current density time course extraction. Regions hMT+ and LOC were defined by motion and object selectivity localizer scans (Huk et al., 2002). V1, hV4, and V3a were defined by standard retinotopic mapping techniques (Engel, Glover et al., 1997).

Definition of regions of interest (ROIs)

For all participants, functional magnetic resonance imaging (fMRI) scans were collected on a 3T Siemens TIM Trio scanner located at the UCSF Neuroscience Imaging Center using a 12-channel whole-head coil and standard Siemens EPI functional imaging sequences with a resolution of 1.7 × 1.7 × 2 mm. The general procedures for these scans (head stabilization, visual display system, etc.) are standard and have been described in detail elsewhere (Brewer, Liu, Wade, & Wandell, 2005; Wade & Rowland, 2010). Retinotopic field mapping produced regions of interest (ROIs) defined for each participant's cortical area V1, V3A, and V4 in each hemisphere (DeYoe et al., 1996; Engel, Glover, & Wandell, 1997; Tootell et al., 1997; Wade, Brewer, Rieger, & Wandell, 2002). ROIs corresponding to each participant's MT homologue “hMT+” were identified using low contrast motion stimuli similar to those described by Huk, Dougherty, and Heeger (2002). Area “lateral occipital cortex” (LOC) was defined based on another standard localizer first described by Kourtzi and Kanwisher (2000). ROIs V1, V3a, hMT+, and LOC were chosen to cover single visual area “clusters”: groups of retinotopic fields with a common fovea (Wandell, Brewer, & Dougherty, 2005). Area hV4 is technically part of the V1 cluster but is relatively distant from the majority of the V1 ROI and, instead, borders a set of foveal representations (VO1 and 2; Brewer et al., 2005) on the ventral surface. It is generally considered to be a classic “ventral stream” area. 

ROI-based signal extraction and preprocessing

Because our regions of interest were defined on the same cortical models as we used for our source imaging procedure, we were able to identify and average cortical current density (CCD) time courses from all mesh points falling within areas V1, V3A, hV4, LOC, and hMT+ in each individual subject. We have shown previously that the spatial resolution of this technique is sufficient to isolate signals from the cortical neighborhood around these ROIs with very little crosstalk (<20%) between regions (Lauritzen et al., 2010). Individual time courses were averaged across all participants to generate a grand average time course. Univariate significance testing was performed in the statistical analysis package SAS (SAS Institute, Cary, NC) with additional statistical analysis in “R” (http://www.r-project.org/) and SPSS (SPSS, Chicago, IL). A univariate repeated-measures ANOVA (with Huyhn–Feldt adjusted degrees of freedom) was used to test for differences on the amplitude of the steady-state visual evoked potential (SSVEP) power at both 6 Hz and 12 Hz: the stimulus flicker frequency and its second harmonic. We also used a univariate test to examine the effect of attention on signal response phase (latency) at these two frequencies. We tested for phase changes in response to two factors (attention and contrast) using circular statistics versions of the t-test where data are modeled with von Mises rather than normal distributions. The parametric statistics was either F-test (large distribution concentration Kappa > 2) or chi-square test (small distribution concentration Kappa < 2). 

Signal normalization

Overall, we find that S-cone stimuli elicit significantly stronger responses per unit cone contrast across the entire cortex compared to luminance stimuli. At first glance, this is puzzling because fMRI and psychophysical measures of cortical and perceptual sensitivity suggest that this situation should be reversed: S-cone stimuli are invariably weaker both perceptually (Dougherty, Press, & Wandell, 1999) and in terms of the population fMRI signals they elicit (Engel, Zhang, & Wandell, 1997; Liu & Wandell, 2005; Wandell et al., 1999) per unit cone contrast. 

The high amplitude of the S-cone SSVEP to extended targets is well documented (Highsmith & Crognale, 2010; Rabin et al., 1994) and is generally attributed to the relative purity of the S-cone waveform: Luminance stimuli drive both parvocellular and magnocellular pathways (Lee, Pokorny, Smith, Martin, & Valberg, 1990) and have multiple input sites in the primary visual cortex (Sincich & Horton, 2005). The VEPs elicited by these stimuli are, therefore, derived from many different neural populations and these individual responses can sum and cancel to generate complex waveforms. In addition, responses from the magnocellular system are relatively transient and may not generate a robust first harmonic in an on–off stimulus presentation paradigm (Kulikowski et al., 1997). In comparison, signals initiated in the S-cones are almost entirely confined to the koniocellular pathway and tend to project to a single layer in V1 (Lachica & Casagrande, 1992; Martin, White, Goodchild, Wilder, & Sefton, 1997; Sincich & Horton, 2005). For this reason, they appear to generate a relatively simple, sustained response that results in a robust first harmonic (see Figure 3). 

Figure 3

 

Cross-subject averages of single-cycle steady-state VEPs in area V1 generated by (a) a 40%, 2-cpd, 6-Hz on–off S-cone contrast grating and (b) a 40% achromatic luminance contrast grating with identical spatiotemporal properties. The S-cone waveform is an almost perfect sinusoid while the luminance response is lower in amplitude and more complex. (c, d) The frequency spectra of the two responses. The amplitudes of the lower harmonics of the S-cone responses are larger, but the luminance-driven signal contains significant responses at higher harmonics. In this paper, we facilitate comparison across ROIs and attentional states by normalizing chromatic and luminance signals to their respective response amplitudes in V1.

In short, the absolute amplitudes of the chromatically and achromatically driven cortical current density time courses are not directly comparable. To facilitate a direct comparison between attentionally driven signal changes in these pathways, we adopt the approach used by Appelbaum, Wade, Vildavski, Pettet, and Norcia (2006) and normalize all chromatic and luminance-driven responses to a single, well-defined response amplitude in the unattended condition measured in V1. To avoid saturation effects, luminance responses were normalized to V1 luminance amplitudes at 20% stimulus contrast and S-cone responses were normalized to V1 S-cone amplitudes at 40% contrast although our results did not differ significantly if we chose to normalize to the highest contrast levels instead. In this way, we were able to compare two types of changes: (1) Relative changes in within-channel signal amplitude generated by variations in stimulus contrast and attentional state as well as (2) differences in chromatic and achromatic responses in different visual areas. 

Parameter fitting

In our modeling section, we fitted contrast vs. response data using a hyperbolic ratio function R = R _max

$\frac{c^n}{c_{50}+c^n}$

with two free parameters: c ₅₀ (the semisaturation constant) and R _max (a multiplicative scaling of the output). The other free parameter in this equation (n, the response exponent) was fixed at a value of 2, which is standard for population response modeling of this kind (Boynton, Demb, Glover, & Heeger, 1999; Chirimuuta & Tolhurst, 2005). In general, our data were well fit by these equations. Parameter significance was tested using a bootstrap procedure (Efron & Tibshirani, 1993). We fit separate hyperbolic ratio functions to 10,000 synthetic mean contrast response data sets generated by resampling (with replacement) the complex response amplitudes from all subjects. We then computed the difference between the fitted values of R _max and c ₅₀ in the attended and unattended conditions. The distribution of these bootstrapped difference parameters allowed us to compute the probability that the true difference between the parameters in the attended and unattended conditions was greater than zero. 

Results

Amplitude changes

Figure 4 shows the response-versus-contrast (RVC) curves for the normalized first harmonics (6 Hz) of the response in all areas under both types of attentional conditions. Most areas exhibited monotonic contrast-dependent changes in response amplitude for both S-cone and achromatic luminance stimuli. In the univariate ANOVA, we found robust amplitude increment with increasing contrast in both luminance (significant in V1/V3A/hMT+, F(3, 30) values ≥ 7.38, p values < 0.01) and S-cone (significant in V1/V3A/LOC, F(3, 30) values ≥ 11.93, p values < 0.01). Area hMT+ showed little response to S-cone stimuli in general. 

Figure 4

 

Normalized response amplitudes for first harmonic component of (a) luminance- and (b) S-cone-driven responses in five visual areas. Responses to attended stimuli (contrast discrimination task) are shown in black; responses during a central letter discrimination task are shown in gray. Error bars are ±1 SEM.

How did attention change the scalar amplitude of responses to the luminance and S-cone stimuli? In the univariate ANOVA, we found little evidence of attentional modulation of S-cone amplitude responses in any area (F(1, 10) values ≤ 3.8, p values ≥ 0.08) but a borderline significant effect of attention contrast on luminance response amplitude in area V1 (F(1, 10) = 4.41, p ≤ 0.06). Further analysis using individual t-tests revealed significant attentionally driven differences for luminance but not for S-cone responses in V1 at high contrast levels (20%, 40%; t ₍₁₀₎ values > 2.3, p values < 0.05). This large amplitude difference at high contrasts hints strongly at the type of gain control to be expected in our modeling section (see below). No other areas exhibited statistically significant changes in response amplitude with attention. 

Phase changes

Steady-state VEP signals have a temporal phase as well as an amplitude at any given harmonic. At a single neuron level, it is well known that increasing contrast leads to a reduction in response latency and, therefore, a phase advance in the response timing for certain neural populations and this phase change is also observed regularly in population measures (e.g., Crognale, Switkes, & Adams, 1997; Di Russo & Spinelli, 1999a; Rabin et al., 1994). The phase advance can be modeled by a normalization stage that alters the average neural membrane time constant depending on the local contrast (Carandini & Heeger, 1994; Carandini, Heeger, & Movshon, 1997). 

Here, we examine the effect of both contrast and attentional modulation on the phases of our chromatic and achromatic SSVEP signals. 

Contrast-driven phase changes for both luminance and S-cone stimuli are shown in Figures 5 (first harmonic) and 6 (second harmonic). We asked the same questions of the phase signals that we asked for the amplitude components: How did attention and contrast change the phase of luminance and S-cone-driven responses? 

Figure 5

 

Contrast and attention-dependent phase changes in the second harmonic (2F) responses of (a) luminance and (b) S-cone responses. Robust contrast-dependent phase advances are evident in most areas for both achromatic and S-cone stimuli. No statistically significant attentional modulation (difference between black and gray lines) is measurable.

Figure 6

 

Contrast-dependent phase changes in the first harmonic (1F) responses of (a) luminance and (b) S-cone responses. Responses to attended stimuli are shown in black; unattended responses are shown in gray. In V1, increasing contrast leads to a significant decrease in response phase for the S-cone but not the luminance-driven responses. Contrast-dependent phase advances are evident in other areas. The small phase difference between the attended and unattended response phases for achromatic targets in V1 is statistically significant (p < 0.05). Phase data are “unwrapped” across the 0/2π boundary.

Again, we found little evidence of attentional modulation of S-cone phase responses (both first and second harmonics) in any area (p values ≥ 0.12 NS) but a significant effect of attention on the phase of the first harmonic of the luminance-driven response in area V1 (F value = 4.83, p < 0.05). In this area, the presence of attention generated a small but significant phase delay of approximately 5 ms. 

Interestingly, we found no evidence for contrast-driven phase changes in V1 at the first harmonic of our luminance data responses (F value = 1.33, p = 0.27) but a strong effect at the second harmonic (12 Hz; chi-square = 62.76, p < 0.01), suggesting that only the transiently responding luminance-sensitive neurons in this region experienced significant contrast gain control. We did find evidence of decreasing first harmonic response phase with increasing contrast in two other visual areas beyond V1: V3a and hMT+ (chi-square values ≥ 14.07, p values < 0.01). For the second harmonic of luminance phase data, the significant contrast effects were found in all visual areas except hMT+ (chi-square values ≥ 12.83, p values < 0.01). 

For both first and second harmonics of S-cone phase data, we found a general and significant phase decrement with increasing contrast in all visual areas (p values < 0.01). Note that the contrast-driven phase changes in the 1F and 2F S-cone signals correspond to approximately equal time delays (approximately 40-ms latency decrease over the contrast range tested) suggesting that the effect of increasing S-cone contrast is simply to change the onset time of a waveform with relatively constant shape. In comparison, the contrast-driven 1F and 2F temporal delays for the luminance signal are different in V1 (approximately 0 ms and 40 ms, respectively, over the tested range) indicating that the SSVEP from this region comprises at least two distinct luminance-driven responses, one of which is modulated by contrast while the other is not. 

Model fitting for response amplitude

When the hyperbolic ratio function of the form R = R _max

$\frac{c^n}{c_{50}+c^n}$

is used to fit contrast response data (Albrecht & Hamilton, 1982), the parameters R _max and c ₅₀ have clearly defined meanings: R _max is the maximum response amplitude and changes in R _max correspond to multiplicative “response gain” output modulations. C ₅₀ is sometimes referred to as the “semisaturation constant” and indicates the point at which responses have reached half their maximum. Changes in c ₅₀ reflect a change in apparent input contrast: a form of multiplicative “contrast gain control” (Heeger, 1992). Here, we fit both attended and unattended contrast response functions using the same equation and allow R _max and c ₅₀ to vary to examine the manner in which attention affects luminance- and S-cone-sensitive neural populations in early visual cortex. By repeating this fitting procedure with randomly sampled combinations of our response data, we were able to perform statistical bootstrapping (Efron & Tibshirani, 1993) to gauge the significance of any changes in R _max and c ₅₀ that we observed. 

Using these non-parametric statistical tests, we found significant attentionally driven changes in R _max for luminance responses only. Specifically, we found a significant change in R _max between the “attend contrast” and “attend letter” conditions in both V1 and the LOC (p < 0.04). Attention altered the R _max of the response in V1 by approximately 12% and in the LOC by 11%. There were no significant differences in the c ₅₀ of achromatic contrast-driven responses in any of the areas we examined nor did we find any evidence for changes in R _max or c ₅₀ in any of the S-cone-driven responses. 

Although the 12-Hz (second harmonic) luminance responses did exhibit a consistent contrast-dependent phase change (see above), the amplitude responses were not well fit by a hyperbolic ratio function and tended to exhibit “hypersaturation” at high contrast—an effect that we attribute to contrast-driven changes in the shape of the waveform that have the effect of distributing power into higher and higher harmonics as the responses become increasingly transient (Table 1). 

Table 1

 

Bootstrapped probabilities of R _max and c ₅₀ changes due to attention in (a) luminance- and (b) S-cone-sensitive populations. Significant response gain (R _max) modulations were observed in areas V1 and LOC for luminance stimuli. In both cases, attention caused an increase in R _max: an amplification of the population output. No significant attentionally driven changes in either R _max or c ₅₀ were observed for S-cone responses.

	V1	hV4	V3A	hMT+	LOC
(a)
R max	0.0416*	0.1130	0.9346	0.3416	0.0478*
c 50	0.4276	0.1128	0.9558	0.4194	0.1374
(b)
R max	0.3184	0.3951	0.4991	0.2817	0.2913
c 50	0.5197	0.3249	0.6606	0.5199	0.6828

 

Note: Asterisk indicates significant parameter testing (p < 0.05).

Discussion

Main results

Sustained attention changed contrast response functions for both luminance and S-cone stimuli in very different ways. For luminance stimuli, attention increased the amplitude and delayed the response time (larger phase) as early as primary visual cortex (V1). However, attending to S-cone isolating stimuli did little to modulate S-cone-driven contrast response functions or phase lags despite the fact that the task difficulty (and, therefore, the nominal attentional load) was identical in the two tasks. We quantified the changes in the contrast vs. response functions by fitting hyperbolic ratio functions with physiologically interpretable parameters. When we did this, we found that areas V1 and LOC exhibited attentionally driven changes in response gain for luminance stimuli but that no area showed a statistically significant change in fitting parameters for the S-cone isolating stimuli. 

Contrast effects on luminance and S-cone stimuli

Consistent with previous human neural population data from both fMRI (Engel, Zhang et al., 1997; Liu & Wandell, 2005; Wade & Wandell, 2002) and EEG studies (Crognale et al., 1997; Di Russo & Spinelli, 1999a; Porciatti & Sartucci, 1999; Rabin et al., 1994), we observed robust response amplitude increments with contrast for both achromatic and S-cone stimuli. We also observed relatively weak responses to S-cone stimuli in areas typically associated with “dorsal” or motion processing pathways: hMT+ and V3a—an observation that is consistent with several other imaging and single unit studies (Barberini, Cohen, Wandell, & Newsome, 2005; Liu & Wandell, 2005; Seidemann, Poirson, Wandell, & Newsome, 1999; Wandell et al., 1999). 

Although increasing contrast has qualitatively similar effects on S-cone and achromatic response amplitudes, it has differing effects on the response phase. In primary visual cortex, we found that both the first and second harmonics of the S-cone-driven responses experienced a phase advance with increasing contrast (indicating a faster response time). In comparison, the first harmonic of the V1 luminance response did not change phase as the contrast increased, although the second harmonic did. 

Phase changes in the second harmonic of the achromatic response are in agreement with those observed by Di Russo and Spinelli (1999b) who measured very similar contrast-dependent phase changes in the response to counterphase flickering achromatic gratings. Di Russo et al. did not measure responses to on–off stimuli like those used here and, therefore, could not identify the “sustained” onset components that we associate with our 1F responses. Porciatti and Sartucci (1999) compared onset latencies of visual evoked potentials (VEPs) at different contrast levels (6% to 90%) for both luminance and S-cone stimuli. Similar to our findings and those of Rabin et al. (1994), they demonstrated that changing the contrast of S-cone stimuli generated large onset latency differences (130 ms from 6% to 90%). However, over the same contrast range, luminance stimuli exhibited much smaller latency changes (20 ms from 6% to 90%). In our steady-state responses, consistent with previous study (Crognale et al., 1997), we measure smaller absolute timing changes over comparable contrast ranges (40-ms S-cone phase shift between 10% and 80% cone contrast), but the general principle that achromatic signal response latency changes are much smaller than S-cone-driven latency changes appears to be maintained. In higher visual areas, we measured significant contrast-dependent phase changes in the achromatic 1F responses—perhaps because the strict segregation between chromatic and achromatic signals is lost after the early stages of V1 (Sincich & Horton, 2005). 

Effects of attention on luminance and S-cone responses

In 1999, Di Russo et al. used a paradigm that was conceptually similar to ours to demonstrate that signals confined to the parvocellular pathway exhibited robust signs of attentional modulation but that the signature of this attentional change differed from that found with stimuli that preferentially stimulated the magnocellular pathway (Di Russo & Spinelli, 1999b). Specifically, they showed that attention modulated the amplitude of PC signals driven by isoluminant red–green gratings as well as signals generated by achromatic gratings. However, attention also had an effect on the temporal phase of the achromatic signals that was not observed in the parvocellular pathway. 

Our data broadly confirm these results for achromatic signals: We observed attentionally driven amplitude and phase changes for achromatically driven signals. What is remarkable is that we observe no significant changes whatsoever when attention is directed toward S-cone gratings. This is especially surprising in light of the fact that our S-cone isolating stimuli clearly generate robust cortical response and experience both contrast-dependent amplitude and phase changes. Neither the amplitude nor the phase of the S-cone signals were affected by attention in any of the areas examined and it is possible that attention simply does not alter the magnitude or timing of cortical responses to S-cone signals. This is consistent with another study (Highsmith & Crognale, 2010) that measured transient evoked response potentials (ERPs) to S-cone stimulus under attended/unattended conditions and found no attention effects on either amplitudes or latencies. 

Although no statistically significant effects of attentional S-cone modulation were observed in these experiments, we note that the normalized responses shown in Figure 4b do hint at attentionally driven signal increases in several areas including V4, V3a, and the LOC. It is, therefore, possible that increasing the SNR of the measurement technique would reveal small but significant effects in these areas. 

In the cases where we measured significant attentional effects on the amplitudes of achromatic luminance responses, the changes were well described as response gain modulations rather than changes in the semisaturation constant c ₅₀. With reference to recently proposed computational models of attention (Boynton, 2009; Reynolds & Heeger, 2009), this suggests that the size of the attentional “spotlight”—the area of visual field or cortex modulated by an attentional “gain field”—is smaller than the overall stimulus extent. This would be consistent with the large annular stimuli used in this experiment. It would be instructive to repeat these experiments using small stimuli where effects resembling contrast—rather than response gain control—might be expected. 

Finally, the effect of attention on the achromatic signals was to increase the phase lag of the 1F component and, apparently, to delay the steady-state signal by approximately 5 ms. This appears to contradict the hypothesis that attention acts via a contrast gain control mechanism (Reynolds & Heeger, 2009; Treue & Maunsell, 1996) because increasing the apparent signal contrast should reduce response time (as seen, for example, in Figure 6). However, a close inspection of the response waveforms at high contrast in the attended and unattended conditions indicates that a different effect may be responsible. In Figure 7, we show the single-cycle waveforms generated by a 40% contrast luminance stimulus in both the attended and unattended conditions. The effect of attention is to sustain the amplitude of the response between 60 ms and 110 ms within the single cycle causing a lag in the fitted 1F sine wave. Because these experiments were conducted in the steady state, assigning absolute timings to peaks and comparing them with ERP experiments is not possible. However, the latencies of the large positive and negative responses at 20 ms, 60 ms, and 117 ms in our plot are not significantly altered by attention. 

Figure 7

 

V1 CCD time course over a single steady-state cycle of 40% luminance modulation. Signals generated by attended and unattended stimuli are shown in dark and light gray, respectively. Attention acts to sustain the signal between 60 and 110 ms causing a lag of approximately t in the fitted response of the first harmonic. Peak onset times are not affected.

A series of papers by Di Russo et al. (Di Russo & Spinelli, 1999a, 1999b, 2002; Di Russo et al., 2001) have identified phase changes in the SSVEP signal due to attentional modulation of luminance gratings. Although these small changes were characterized predominantly as reductions in signal lag, we note that at lower contrasts (Di Russo et al., 2001) or at low frequencies (Di Russo & Spinelli, 2002) the effects are similar to those we observe here. In addition, Appelbaum and Norcia (2009) recently reported a 6-ms delay in their SSVEP data caused by feature-based attention using a paradigm very similar to ours. 

Our phase data are not consistent with a timing change caused by gain control mechanisms. However, given the small absolute magnitude of the attentionally driven increases in response amplitude that we observe, we would not expect to be able to resolve the phase changes associated with attentional gain control. Instead, we propose that the sustained response we observe in our own data may be driven by a feedback signal to V1 (Aine, Supek, & George, 1995; Martinez et al., 2001; Vidyasagar, 1998, 1999), which could occur in addition to any multiplicative amplitude modulations of the stimulus-driven response (Lauritzen et al., 2010). 

Conclusions

Using electrical source imaging (ESI), we measured the neural contrast vs. response functions for attended and unattended S-cone isolating and achromatic stimuli. Attention generated amplitude and timing changes in neural responses to achromatic contrast in V1 but had no effect on neural responses due to S-cone isolating stimuli. Our results are consistent with an anatomically early cortical site of attentionally driven neuronal gain control. 

Acknowledgments

This work was funded by NIH EY018157-02 and NSF BCS 0719973 to AW. We thank Justin Ales, Tony Norcia, and Preeti Verghese for helpful discussions and support in the preparation of this manuscript. 

Commercial relationships: none. 

Corresponding author: Alex R. Wade. 

Email: [email protected]. 

Address: Smith-Kettlewell Eye Research Institute, 2318 Fillmore Street, San Francisco, CA 94115, USA.