Now you see it: frontal eye field responses to invisible targets
John Assad
John Assad is in the Department of Neurobiology, Harvard
Medical School, 220 Longwood Ave., Boston,
Massachusetts 02115, USA. jassad@hms.harvard.edu
Neurons in the frontal eye field respond even when a visual target is perceptually
masked, but small variations in their activity predicts whether a monkey will
respond to the stimulus.
Over the past several decades, neurophysiologists have made significant
progress toward understanding the chain of neuronal processing linking sensation
to action. However, like excavators slowly converging from opposite sides
of a tunnel, their efforts have focused largely on the extremes of the problem—the
initial sensory input and final motor output. For eye movements made to visual
targets, we understand in detail both the early visual analysis occurring
in the retina, thalamus and primary visual cortex and the brainstem mechanisms
controlling eye movement. Yet, the sheer complexity and flexibility in our
responses to visual stimuli argue that numerous stages of neuronal processing
must intervene. On page 283 of this
issue, Thompson and Schall1 provide an intriguing glimpse into
how one likely site of such processing, the frontal eye field (FEF), uses
visual information to control a decision about whether to move the eyes. Under
masking conditions that make the target's visibility unreliable, the authors
report that FEF neurons reliably respond to the target whether or not the
monkey perceives it, but that small variations in the magnitude of their response
predict very accurately whether an eye movement will be made.
David Ferrier first described the FEF in 1875 as the area in frontal cortex
that most readily elicited eye movements when electrically stimulated. Many
lesion and inactivation studies have supported this view, and tracer and stimulation
experiments have revealed strong projections from the FEF to areas involved
in generating saccades, such as the superior colliculus and pontine oculomotor
nuclei2. In addition to these motor functions, the FEF also
seems to be a site of visual integration, in that it receives inputs from
a vast swath of extrastriate visual cortex. The FEF thus seems ideally positioned
at the confluence of visual inflow and motor outflow to mediate visual−motor
interactions. Consistent with this view, single-unit recording in monkey FEF
has revealed neuronal signals related to both sensory stimuli and saccades,
sometimes in the same neuron. In a typical experiment, a spot of light is
flashed in the visual periphery, and the monkey is required to saccade to
the location of the target. Many FEF neurons discharge shortly after the presentation
of the stimulus, and then also around the time of the eye movement3.
Given this dual nature of FEF responses, it seems reasonable that the FEF
might be an important center for integrating visual input to control eye movement—a
sort of decision point or dispatcher.
Thompson and Schall varied the usual saccade task by using backward masking
to make the stimulus sometimes perceptually invisible. If a brief target is
immediately followed by another stimulus (the mask), humans will report that
the target is invisible, and similarly monkeys will behave as though it was
not seen. The effectiveness of backward masking depends on the time separating
the stimulus and the mask (commonly called stimulus onset asynchrony, or SOA):
in general, briefer intervals lead to more effective masking4,
5.
Thompson and Schall trained monkeys to first fix their gaze on a point of
light at the center of a computer screen. On most trials, a dim target spot
was flashed at one of eight locations around the fixation point, one of which
was within the receptive field of the FEF neuron under study; on the remaining
trials, no target appeared (Fig. 1). On all trials,
a ring of bright spots then masked all eight locations. The monkeys' job was
to saccade to the location of the target if it was visible, but to withhold
the saccade otherwise.
Figure 1. Stimuli for the backward-masking protocol.
A dim target appears at one of eight positions on some trials, but not
on others. On all trials, eight bright masking stimuli then appear, which
can prevent the monkey from perceiving the target.
Rather than masking the target completely, the authors lengthened the SOA
to reduce the effectiveness of the mask. Under these conditions, humans (and
presumably monkeys too) find that their perception of the target becomes rather
unreliable: on some trials they will report seeing it, whereas on others they
will swear it was not shown. The authors adjusted the SOA so that when the
target was actually present, the monkeys only made saccades on about half
the trials. The beauty of this approach is that an identical visual stimulus
can be examined under conditions in which the stimulus does or does not reach
perceptual awareness—assuming that the animals must be aware of the
target to respond to it. By factoring out the physical visual stimulus on
the retina in this manner, one can then look for neuronal responses that correlate
with the animal's perception. This approach has been applied recently with
great success in more 'purely' visual cortical areas, using perceptually bistable
visual stimuli such as binocular rivalry6,
7,
8 or ambiguous
structure-from-motion displays9. In general, as the signal passes
to successively higher visual areas, an increasingly higher proportion of
neurons show activity that correlates with the animal's report of its perception.
This is in line with the emerging view that higher visual cortical areas 'filter'
the visual scene to highlight inputs that are relevant to the behavioral task
and discard distracting information10. Under this view, we would
predict that a large percentage of FEF neurons, if not all, should reflect
the monkey's perception, and thus that the FEF neurons recorded in Thompson
and Schall's experiment would not respond much when the monkeys reported that
the target was not visible.
Instead, rather surprisingly, almost all FEF neurons responded strongly
to the flashed target regardless of the monkey's subsequent report. This result
raises several interesting issues. First, it constrains the possible neuronal
mechanisms underlying backward masking. In contrast to proposals that masking
is mediated by the early visual system, perhaps even by the retina, the finding
that perceptually masked stimuli can evoke strong visual responses in the
FEF argues that masking is not exclusively a function of early vision. Second,
this result indicates that the convergence of visual inputs allows FEF neurons
to respond to visual stimuli that do not guide action, even though many of
the extrastriate visual areas providing input to the FEF probably respond
much less well to stimuli that do not reach perceptual awareness. This implies
a more broadly parallel visual influence on the FEF than had been appreciated.
It also raises the question of just how much further sensory signals are conveyed
in the motor chain. Visual signals have been identified in the superior colliculus,
but might they also reach the oculomotor nuclei, and if so, what would they
do there?
The authors also noticed that although most FEF neurons responded strongly
to the target under all conditions, the response was larger on trials when
the monkeys reported seeing the target. This difference in activity was slight
(about 30% on average), but extremely consistent among the population of neurons.
This suggests that even though both unperceived and perceived visual signals
reach the FEF, the FEF could in principle distinguish the two. Stated more
strongly, FEF neurons might actually make the decision of whether the stimulus
will be perceived, based on small variations in the size of the response to
the stimulus. If this interpretation is correct, it would be remarkable. One
possible caveat is that FEF neurons are also active during eye movements,
and the monkeys signaled their perception with an eye movement, so that the
slightly larger responses on target-perceived trials might be more related
to dumb motor function than to decision-making. However, because the target
responses were of such short latency, and the monkeys could not predict the
location of the target in advance, this interpretation seems unlikely. Thus
the tight correlation between the size of the target response in the FEF and
perception, coupled with the importance of the FEF for eye movements, suggests
that the FEF may be critically involved in deciding whether to move the eyes.
For some neurons, on trials in which the animal made a saccade to the location
of the neuron's receptive field, the background activity before the target
onset was slightly elevated compared to identical trials in which the animals
maintained fixation. These small variations may have arisen randomly and then
biased the upcoming behavior, acting like a tie-breaker. One worry is that
the baseline increase merely reflected an occasional bias by the animal to
saccade to that location, but because the monkey could not predict the target
location, any behavioral bias should have been averaged away. Another caveat
is that such a razor-thin influence may only affect near-threshold decisions,
when the animal is teetering on the brink of whether to respond. However,
near-threshold choices probably arise quite often during natural viewing,
for instance, under conditions of low illumination.
These results raise a host of intriguing questions. Is the decision to
move the eyes actually sorted out in the FEF, or is it passed along to another
downstream motor area? Is the FEF only involved in perceptual decisions that
result in eye movements, or is the signal more generally accessible? Assuming
that the decision depends on the activity of many neurons, doesn't the close
relationship between single-unit responses and perception argue that the responses
of many neurons should be correlated from trial to trial? Whatever the answers,
it seems a good bet that the FEF will continue to provide a useful model for
examining how the brain makes up its mind.
