Electrical neuroimaging evidence that spatial frequency-based selective attention affects V1 activity as early as 40-60 ms in humans

Neurons in the primary visual cortex (area V1) not only code simple features but also whether image elements are attended to or not [2]. However, the timing of such attentional modulation is not well understood and there is no agreement in the literature. For example, it is believed that when they are not altogether absent [3–5], attentional effects are weaker [6, 7] in V1 than in associative visual areas [7–9] and occur after an additional delay. Several fMRI studies have reported robust effects of attention on V1 responses, demonstrating that attentional selection operates very early in the visual pathway [10]. According to Pessoa and coworkers [11], V1 attention effects are sometimes observed with fMRI but not with other techniques because they do not take place during the initial stimulus-related response (60-90 ms), as shown by combined fMRI/EEG and MEG studies [12], but at longer latencies in the time range 150-250 ms. According to these authors, it appears that V1 is "reactivated" in the 150-250 ms post-stimulus time range within the focus of spatial attention [8], as a sort of re-entrant feedback. In addition, some data from electrophysiological and fMRI recordings clearly indicate that spatial selection is faster and more effective than feature-based attention [13, 14].

As for the timing of attentional modulation, the ERP technique is a valuable investigative tool because of its optimal temporal resolution, which, combined with modern source reconstruction techniques (such as LORETA [15]), can be informative about both the time course of attentional selection and the identification of underlying neural structures.

Indeed, while it is likely that space-based attention can modulate the extra-striate and even striate visual cortex [1, 16] at a very early latency [17], numerous ERP studies have attempted to identify the stage at which attention can select non-spatial properties of the stimulus without reaching clear converging evidence. Harter and Previc [18] first attempted to assess the effects of selective attention on the activity of cortical size channels, finding an increase in amplitude of the selection negativity (SN) as early as 160 ms to attended check sizes. More recently, an ERP study with checkerboard stimuli [19] showed that the earliest signs of selective attention to check size consisted of an occipital selection negativity (OSN) at about 140 ms and a frontal selection positivity (FSP) somewhat earlier, at 120 ms. Similarly, in two studies involving selective attention to spatial frequency [20] and a given conjunction of spatial frequency and orientation [21], an attention-related anterior positivity was found at about 120 ms latency; this preceded a posterior selection negativity just after 200 ms.

A different ERP investigation of selective attention for low vs. high spatial frequency [22] showed that while attended high spatial frequency stimuli elicited an early negative difference potential (ND120) starting at about 100 ms, attended low spatial frequency stimuli elicited a positivity (PD130) in the same latency range. The neural sources of both effects were estimated by dipole modelling to lie in the extrastriate visual occipital areas, while the C1 component (60-100 ms) generated in the striate (and extrastriate) cortex was not affected by attention to spatial frequencies, according to the authors.

On the other hand, a number of other ERP studies have identified earlier P1 selective attention effects for object-based characteristics such as colour [23], colour and movement [24], check size and grating spatial frequency [25–27], transparent motion [28] and orientation [29]. Furthermore, Proverbio and Zani recently found spatial-frequency [30, 31] and shape-based [32] attentional modulations of the P/N80 (C1) component as early as 60 ms, and interpreted this finding as a clear indication of striate modulation.

The possible reasons for these differences among findings and inconsistencies in the ERP literature have been widely discussed elsewhere [32] and range from large inter-individual variability in the morphology of VEPs and their attention effects, to a variety of methodological factors including differences among studies in stimulus presentation rate (e.g. too fast), inter-stimulus interval (e.g. too short: 200-500 ms in [9]; 300-550 ms in [22]), contrast and luminance levels, duration (e.g. too short: 50 ms in [9]), excessive filtering or adjar adjustments.

The overall aim of the present experiment was to investigate this matter further by using high density EEG recording from a large number of right-handed viewers engaged in a selective attention task requiring them to attend to both spatial frequencies and locations of gratings. In order to optimize the signal to noise ratio (and pick up the tiny V1 attentional modulation), large numbers of subjects (twenty-nine) and of trial repetitions were used in this study (13,312 stimulus repetitions per subject). In addition, swLORETA source reconstruction techniques were employed on the difference-waves of interest to identify the neural bases of spatial frequency-based attention effects.

It was expected that selective attention would affect early-latency responses at the level of the C1 response, a component of the VEP considered to indicate the initial afference of retinotopic regions in the human visual cortex (V1) [33], with an onset over the central parieto-occipital scalp between 45 and 60 ms. A similar outcome was recently found for the attentional modulation of Gabor pattern luminance [16].

In this paradigm, gratings of different spatial frequencies were randomly presented in the upper and lower quadrants of the visual field in a task requiring conjoined/simultaneous attention to spatial location and spatial frequency. Visual evoked potentials (VEPs) showed the usual effects of spatial frequency and retinal position on the amplitude of sensory components, with larger N80 responses to upper stimuli and to higher spatial frequencies. P/N80 amplitude was greater at mesial occipito/parietal sites and inverted its polarity as a function of the above factors. As often reported, the later P120 response, greater at lateral occipital sites, had a larger amplitude to low spatial frequency gratings and showed the strongest spatial attention effects. Overall, these effects are rather canonical and are well documented in the literature [36, 37]. Equally well known is the striate origin of the C1 component of VEPs, demonstrated by both electrophysiological and neuroimaging techniques [33, 38–41].

The P/N80 inversion as a function of stimulus horizontal meridian and spatial frequency is highly consistent with previous electrophysiological literature [26, 31, 37, 41, 31, 43, 44].

ANOVA performed on the mean amplitude value of the C1 component in the first time window considered (40-60 ms) showed significant frequency-relevance effects at both attended and unattended locations at right hemispheric sites, and at the attended location at the left hemispheric sites. In addition, frequency-relevance effects at the attended location were larger for LVF gratings. This phenomenon might be interpreted in two ways. One possibility is that selective attention to spatial frequency exhibited the renowned hemispheric asymmetry for frequency processing, the right hemisphere being more efficient in processing the range of low (0.75 c/deg) than high (6 c/deg) spatial frequencies [45–47]. The other hypothesis is that the LVF/right hemisphere advantage might reflect an early low-level sensory bias for visual processing. Indeed, there is evidence of a similar LVF advantage for the processing of simple visual stimuli in simple RTS paradigms [48]. In any case, the matter deserves further investigation. The right hemispheric and LVF attention effects at the earliest stage of visual processing are strongly consistent with LORETA source reconstruction indicating an attentional effect for the relevant frequency (F+ vs. F-) in the right occipital cortex (BA17) in both C1 time windows (40-60 and 60-80 ms).

The early onset of the spatial frequency-based attention effect is compatible with the most recent findings on the timing of space-based attentional selection, e.g. [16]. In addition, the early (40-60 ms) emergence of robust 6 but not 0.75 c/deg frequency selection effects are compatible with recent findings [33] showing that at high contrast levels, the parvocellular system makes the biggest contribution to generating the C1 component starting at about 45 ms. Overall, evidence of stronger frequency relevance effects for high (6 c/deg) than low (0.75 c/deg) spatial frequency gratings has previously been reported in similar ERP attentional studies [22, 25, 26, 31, 32]. This inhomogeneity may be ascribed to the difference in contrast sensitivity across spatial frequency ranges, with 4-5/deg spatial frequency bandwidth being the optimal range for the human visual system [31, 49, 50]. In this light, it is conceivable that the earliest target/non-target effect might be observed in V1 for the frequency band eliciting the most optimal response among V1 neurons (6 rather than 0.75 c/deg). The preference for 6 over 0.75 c/deg gratings is also supported by behavioural data, showing faster RTs to the former stimuli.

The interaction between location relevance × frequency relevance, observable from the earliest sensory stages, suggesting stronger attentional selection effects at the attended location, is compatible with previous ERP literature [26, 31, 32] suggesting similar effects as early as 60 ms post-stimulus. The mechanism subserving this attention enhancement is probably related to the mechanism by which covert spatial attention increases contrast sensitivity via contrast gain, thus enhancing spatial resolution, described in neurophysiological and psychophysical studies [51, 52].

These data strongly influence the existing assumptions and models of selective attention according to which the effects of attention on V1 activity take place not during the initial stimulus-related response (60-90 ms) but, instead, at longer latencies in the time range 150-250 ms, as a sort of re-entrant feedback [8, 11, 13].

The present data firmly establish that, indeed, as a result of task attentional relevance, visual cortex responsivity (including V1 activity) is cued to enhance/improve the processing of the attended spatial frequency, at both attended and unattended locations. While later (P1) frequency-relevance effects were stronger at the attended location (L+F+ vs. L+F-), the earliest frequency relevant effects, namely C1 modulation between 40 and 100 ms (see Table 1), exhibited strong frequency-relevance effects per se (F+ vs F-) (see Fig. 6). These data support the hypothesis that object-based selective attention processes might also be carried out at the earliest processing stage within the striate visual cortex, similarly to what was found for spatial attention most recently [16]. Indeed Kelly and coworkers employed a visuo-spatial task in which subjects were cued on each trial to direct attention toward 1 of 2 locations in anticipation of an imperative 6 c/deg Gabor stimulus and were required to detect a region of lower luminance appearing within the Gabor pattern 30% of the time at the cued location only. The data show a clear spatial attentional enhancement of the C1, beginning as early as its point of onset (57 ms), which inverted in polarity as a function of upper vs. lower hemispace. Source analysis of the attentional modulations pointed to generation in striate cortex.

It's interesting to note that, in our study, C1 attention effect did not invert in polarity as a function of quadrant of stimulation (as expected on the basis of P/N80 reversal to upper vs. lower stimuli). In fact, while location relevance did not affect much of the earliest sensory processing, and later on it enhanced the positivity of VEPs to gratings falling at the attended location, frequency relevance increased the negativity of both C1 and P1 responses regardless of quadrant of presentation. The presence of this attentional modulation, a sort of early selection negativity (SN) [22, 25] that subsequently enhanced the amplitude of posterior N1 and N2 components (as clearly visible in Fig. 3 and 5), supports the hypothesis that C1 might index the activity of multiple generators beyond primary visual cortex.

These findings are paralleled by a number of electrophysiological data suggesting several sources for the early VEP based both on pathological [53] and control data. In addition, MEG findings [54, 55] have demonstrated the involvement of V1, V2, V3, inferior and superior lateral occipital gyri and intraparietal sulcus in generating post-synaptic potentials in the 70-100 ms post-stimulus time window.

As for more anterior brain areas, in our study the frequency-based attention-related activation of BA6, BA45/46 and BA10 prefrontal areas was quite small in the early phase of C1 (below 0.7 nA of magnitude) but became much stronger and reliable (6-9 nA) in the next time window (60-80 ms): This pattern of results is consistent with the electrophysiological and SCD mapping data provided by Foxe and Simpson [56] showing an early activation of dorsolateral prefrontal cortex in the C1 range (as early as 80 ms) during a cued multisensory attention task. At this regard it should be considered the crucial role of the frontal lobe in spatial attention allocation, which may occur even before V1 response. It is for example known that the frontal eye field has neurons that discharge before visually guided saccades [57] thanks to corollary discharge signals coming from superior colliculus pathway and travelling through mediodorsal thalamus to the frontal eye fields, in the prefrontal cortex [58].Supporting evidence comes also from TMS studies showing the involvement of both frontal eye fields [59, 60] and dorsolateral prefrontal cortex [61] in the early modulation of visual cortex during covert voluntary attention tasks.

Indeed, the direct role of the frontal lobe in modulating visual processing and particularly the V1 response has been demonstrated. For example it has been shown that single pulses of transcranial magnetic stimulation (sTMS) restricted locally to frontal cortical areas enhance visual perception of phosphenes and flashed alphabetical characters [62]. According to the authors, the anterior frontal lobe can gate information from primary visual cortical areas leading to enhanced perception through its powerful connections with the thalamic intralaminar system. It has been proposed that the frontal-lobe projections to the thalamic intralaminar nuclei can selectively enhance sensory processing by the primary cortical receiving area, thus giving rise to the early attentional modulation of V1.

It is quite interesting to consider at this regard that, in humans, activity of thalamocortical circuitry is reflected by gamma activity in the EEG [63, 64] and indeed there is clear evidence of both beta and gamma synchronization around the time of C1, beginning around 50 ms. The oscillatory data suggest the possibility of long distance synchronization as an explanation of early V1 effects. Besides hard-wired anatomical pathways which could convey information to the occipital cortex at short latency, long distance synchronizing effects of attention on V1 neurons should be also be considered.

As for the potential limitations of this study it may be considered that fitting the total time period with ICA methods may have strengthened or weakened the conclusion of latency linked attention in V1. Further investigation will be able to shed some light on this matter.

In conclusion, the present data highlight the limitation of the current model of object-based visual selective attention in demonstrating that visual cortex responsivity (including V1 activity) is cued to enhance/improve the processing of attended objects at the earliest sensory level (C1).