Dissociating prediction and attention in the brain: a combined TMS-EEG study

It is thought that one of the fundamental organization principles of the brain
is the ability to predict upcoming events in the environment (Rao and Sejnowski
2002; Friston 2005; Bar 2007; Enns and Lleras 2008). For example, when you grab
a cup of coffee, your brain predicts how your hand is moving through space, and
how the coffee cup will feel when you have grabbed it (Blakemore, Goodbody et
al. 1998; Wolpert and Ghahramani 2000). When the coffee cup feels different
than you had expected (e.g., cold when you thought the coffee was still hot),
you immediately notice something isn*t quite right. Predictions thus allow the
brain to ignore predictable (and therefore uninteresting) events in the
environment, while enhancing the saliency of new, unexpected events. Using
prior knowledge to predict upcoming events also has substantial computational
advantages, effectively constraining the vast amount of incoming sensory
information by our predictions.
Predictive mechanisms have been observed at many stages of neural information
processing, ranging from the retina (Srinivasan, Laughlin et al. 1982; Hosoya,
Baccus et al. 2005) and primary sensory cortices (Sharma, Dragoi et al. 2003;
Naatanen, Jacobsen et al. 2005) to higher-order perceptual regions
(Summerfield, Trittschuh et al. 2008). In fact, many neurocognitive phenomena
like the suppressed activity for expected simple visual (Summerfield,
Trittschuh et al. 2008), auditory (Naatanen, Jacobsen et al. 2005) or
somatosensory (Blakemore, Frith et al. 1999; Voss, Ingram et al. 2006) events,
and the enhanced activity for unexpected linguistic events (Kutas and Hillyard
1980; Hagoort, Hald et al. 2004) can all be seen as activity modulations
related to prediction and prediction error.
Although predictive mechanisms appear to play a large role in shaping our
perception, cognition and action, it is largely unclear how such mechanisms are
implemented in the brain. A recent framework - known as *predictive coding* -
provides a good working hypothesis. In this framework, it is proposed that
prior knowledge allows the generation of expectations at multiple hierarchical
levels in the neocortex (Mumford 1992; Hawkins 2004; Friston 2005).
Expectation-related information (predictions) is fed back to preceding cortical
areas, such that feedforward sensory input is interpreted at each cortical
stage within the context of the prior expectation (Rao and Ballard 1999; Rao
and Sejnowski 2002) (see Figure 1). This framework has recently received
increasing empirical support (Summerfield, Egner et al. 2006; Garrido, Kilner
et al. 2007; den Ouden, Friston et al. 2008).

One key hypothesis of predictive coding is that higher-order regions in the
cortical hierarchy try to predict their input via feedback connections (Hawkins
2004; Friston 2005). In this project, we will test this hypothesis by assessing
feedback connection strength as a function of prediction and attention.
Subjects will be trained to expect either faces or objects on a trial-by-trial
level by a predictive cue. Moreover, they will have to make a judgment on one
of the categories, thereby manipulating attention. Just before presentation of
the stimulus, a low intensity TMS pulse will inject current in the occipital
face area (OFA, a higher-order visual area involved in the representation of
faces)(Pitcher, Charles et al. 2009) or the lateral occipital complex (LOC, a
higher-order visual area involved in the representation of objects) during
various levels of anticipation and attention to motion, and the consequence of
the injected current on downstream areas will be registered using EEG
(Morishima, Akaishi et al. 2009). This allows us to trace signal transmission
from higher-order to lower areas in the cortical hierarchy as a function of
prediction and attention.

The main objective is to test the hypothesis that predictions are implemented
in the human brain by modulating the strength of backward connections from
higher-order to lower-order areas. This modulation of effective connectivity
will be assayed by investigating TMS-locked evoked potentials in the occipital
cortex.
The secondary objective is to dissociate the relative contribution of
prediction and attention, two key forms of top-down influence on visual
processing, within the same paradigm. We will be able to orthogonally
manipulate these two factors by varying predictability and relevance of the
stimuli.

The study is designed as a crossover experiment with healthy adult volunteers.

The study will investigate the role of LOC and OFA during expectation of visual
stimuli of faces and objects. Stimulation of both regions will be compared to
vertex stimulation. We expect that stimulation of LOC and OFA will lead to
larger visual evoked potentials during anticipation of objects and faces,
respectively. This effect is expected independently of the relevance of the
stimuli. Furthermore, we expect a modulation of TMS-induced visual evoked
potentials by relevance: attending to objects/faces will lead to a larger
TMS-induced potential over LOC/OFA, irrespective of whether this stimulus
category is anticipated.

Subjects will come for a total of four sessions - one session to acquire the
anatomical image of the brain and establish the active motor threshold (see
below), and one session for each of the three targets regions: LOC, OFA, and
vertex.
Prior to the experiment, participants are informed about the study in detail
and about the possible risks. They are screened using a questionnaire to ensure
their eligibility for participation. In the first session of the pilot
experiment, participants will practice with the behavioural task. Afterwards,
we will establish subjects* active motor threshold (aMT) as determined using
TMS pulses. The aMT is defined as the lowest TMS intensity needed to evoke a
reproducible and measurable (with electromyography) muscle twitch in the first
dorsal interosseus of the right hand (Rossini et al. 1994). The aMT provides an
indication of the excitability of a participant*s brain and will be used to
determine the stimulation intensity in the later TMS sessions.
In sessions 2-4, participants receive single TMS pulses at 80% of the
participant*s aMT to LOC, OFA and vertex, respectively. The placement of the
coil will be guided an anatomical MRI scan of the individual subject. The
stimulation locations will be determined using the BrainSight TMS-MRI
co-registration system. This system allows to navigate the TMS coil in relation
to the individual anatomical MRI in real-time with millimetre accuracy. Each
session will consist of 2 expectancies (predicted, unpredicted) X 2 stimulus
types (faces, objects) X 2 attentional levels (attended, unattended) X 50
trials. Each trial is expected to last 5 sec. The participant will therefore
receive 400 pulses to the stimulated site, over a period of 45 minutes,
including several breaks. Sessions 2,3, and 4 are spaced at least 7 days apart
and will be counterbalanced across subjects.

We will use single-pulse TMS during the anticipitory phase of the experiment,
and we will measure brain activity with help of EEG. Earlier studies have shown
the effectivity of this set-up, using low intensity single-pulse TMS (80% of
aMT)(Morishima et al 2008).

The magnetic field used during the experiment are of limited size and short
duration, and there is no indication of an influence on health. Because we will
use single-pulse TMS with a low intensity (80% of aMT), the burden and risks
are negligible.