Studies in Humans

In humans, a large body of neuroimaging research has described top-down effects on metabolic and hemodynamic indicators of neural activity in early sensory cortices across a wide range of experimental paradigms.

Using spatial attention paradigms similar to those employed in numerous EEG stud-ies (Clark & Hillyard, 1996; Gomez Gonzalez et al., 1994; Johannes et al., 1995), Tootell et al. (1998) as well as Gandhi, Heeger, and Boynton (1999) examined the effects of selec-tive attention on V1 activity using fMRI. Both groups observed attentional modulations in primary visual cortex; in the study of Gandhi and colleagues, these modulations were on the order of 25% of stimulus-evoked activity.

Ress, Backus, and Heeger (2000) used a simple contrast-detection paradigm with

Introduction

auditory cueing to examine BOLD responses of retinotopically organized visual cortex as a function of stimulus presence or absence. They observed significant responses in V1 through V3 which were virtually the same whether a stimulus was presented or not.

This ‘base response’ was systematically linked to behavioral performance, with larger base responses in all three areas predicting better performance. Importantly, the base response was retinotopically specific in that it was observed only in visual cortex regions representing the experimental stimulus. Strikingly, these responses were absent at higher contrast levels, where behavioral performance was at ceiling. The authors interpreted trial-to-trial fluctuations in the base response as correlates of fluctuations of attention which in turn affect performance. While its neurophysiological origins remain to be clarified, the base response probably reflects a specific biasing signal recruited during difficult perceptual judgements and thus suggests the operation of top-down effects at the level of early visual cortex. However, as the authors point out themselves, it is impossible from their data to conclude whether the base response originated in V1 and was passed on to higher visual areas or vice versa.

Mirroring the results of animal studies (Ito et al., 1998), it has been demonstrated that perceptual learning in a texture discrimination task (Karni & Sagi, 1991) selec-tively increases neural activity at the representation of the trained stimulus in human V1 (Schwartz, Maquet, & Frith, 2002). Importantly, additional analyses of functional connectivity patterns suggested no involvement of other brain regions in mediating this effect. Similar results have been reported by other groups (Furmanski, Schluppeck, &

Engel, 2004; Sigman et al., 2005). In another study, Schwartz et al. (2005) varied atten-tional load (Lavie, 2005) at fixation and found reduced BOLD responses to irrelevant peripheral distractors under increased attentional demands. This increased attentional filtering was seen even at the level of V1, with patterns of reduced activity conforming to a surround-suppression profile (Bahcall & Kowler, 1999). In both cases, long-range horizontal connections recruited during learning or gated as a function of attentional demands could conceivably mediate the suppression of task-irrelevant stimuli, in accor-dance with the proposals of Gilbert and co-workers (e.g. Gilbert & Sigman, 2007).

Using concurrent transcranial magnetic stimulation (TMS) and fMRI, Ruff et al.

(2006) provided direct causal evidence for top-down modulations of early visual cortex activity. In their study, frontal eye-field TMS led to a characteristic pattern of increased BOLD responses throughout V1 to V4 for regions representing the peripheral visual field. Activity in central regions, on the other hand, was reduced by TMS and this pattern of results was observed both in the presence and the absence of visual stimu-lation. Importantly, a separate psychophysical experiment showed behavioral effects in the same direction, i.e. contrast thresholds were decreased for peripherally presented stimuli (Gabor patches) during TMS. Thus, changes in higher cortical areas can mod-ulate neuronal responses in early sensory cortices and these modulations can in turn modify conscious perception even of very simple stimuli. Similar results have also been reported by Taylor, Nobre, and Rushworth (2006) using combined EEG and TMS .

While these studies strongly suggest high-level influences on activity in V1, the de-lay of the hemodynamic response does not permit to draw strong conclusions about

Introduction

the time-course of many of the reported effects. However, numerous studies have also demonstrated effects of selective attention on pre-stimulus activity in early sensory cor-tices, suggesting that top-down mechanisms may exert relatively long-lasting influences on early sensory cortices and thereby influence the latter’s response characteristics.

Kastner, Pinsk, De Weerd, Desimone, and Ungerleider (1999) observed BOLD sig-nal modulations in terms of both baseline shifts and attentiosig-nal modulations of visually evoked responses in retinotopic visual areas. Both classes of effects differed as a func-tion of simultaneous or sequential presentafunc-tion of visual stimuli, as described previously by the same group (Kastner, De Weerd, Desimone, & Ungerleider, 1998). Importantly, retinotopically specific increases of neural activity in preparation for peripherally pre-sented stimuli was observed even at the level of V1, although this effect was less reliable than in higher visual areas (see also Schwartz et al., 2005). Stimulus-evoked responses were modulated by attention only from V2 onwards, although this may be linked to the large extent of the stimuli, potentially reducing the involvement of V1 with its small receptive fields in attentional biasing.

O’Connor, Fukui, Pinsk, and Kastner (2002) observed attentional modulations of both stimulus-evoked responses and baseline activity in the lateral geniculate nucleus (LGN), the principal thalamic relay of the visual pathway connecting the retinae to visual cortex. Thus, selective attention could conceivably gate information processing even at subcortical stages (Crick, 1984), indicating a possible physiological mechanism for attentional modulations of early stimulus-evoked activity in primary sensory cortices.

Muckli, Kohler, Kriegeskorte, and Singer (2005) used an apparent motion paradigm, real motion stimuli, and retinotopic mapping to examine the neural representations of apparent motion paths. They observed equivalent activity patterns in V1 both for apparent and real motion. Just as the findings of Shuler and Bear (2006) reviewed in the preceding section, these results are related to the idea that activity in early sensory cortices may reflect the matching of dynamic predictions and actual stimulus patterns (Rao & Ballard, 1999). Similar results have been reported by Summerfield et al. (2006), and Bahrami, Lavie, and Rees (2007) later demonstrated that V1 activity in response to invisible stimuli is itself subject to attentional effects.

Ruff and Driver (2006) showed that attentional modulations in visual cortex can be observed not only for target stimuli, but also for irrelevant distractors. In a psy-chophysical experiment, they found that advance knowledge reduced the behavioral costs associated with distractor occurrence. In a subsequent fMRI experiment, they observed changes in early visual cortex activity contralateral to cued distractors, and these changes were already present during the cue-target interval. Their results suggest that top-down influences on sensory processing may not only bias competition in favor of relevant stimuli, but may also serve to suppress neural processing of predictable distract-ing information, similar to the finddistract-ings of Schwartz et al. (2005). However, since Ruff and Driver (2006) did not acquire retinotopic maps in their subjects, the involvement of V1 in these effects remains unclear.

Even more recently, Bestmann, Ruff, Blakemore, Driver, and Thilo (2007) observed enhanced phosphene perception following TMS over visual cortex if subjects paid

atten-Introduction

tion to the stimulated quadrant. Although the specificity of TMS stimulation is limited, these findings suggest that attention can enhance levels of excitability in sensory cortices independent of the source of information, i.e. even if visual cortex is stimulated directly rather than via the retino-thalamic pathway.

The preparatory effects reported in these studies may correspond to non-linear inter-actions between top-down signals and sensory processing as postulated by Hup´e, James, Girard, Lomber, et al. (2001). Although in the latter study, differences in baseline activ-ity were specifically excluded as mediators of the observed modulations of early visual cortex activity following MT inactivation, baseline shifts as a consequence of selective attention have been reported in the awake monkey (Luck et al., 1997). The BOLD signal as a complex, cumulative indicator of neural activity (Logothetis, 2002; Logothetis et al., 2001) could conceivably capture large-scale preparatory processes which may have gone unnoticed in single-cell or multi-unit studies in non-humans.