Continuous processing in macaque frontal cortex during visual search

Abstract

A central issue in mental chronometry is whether information is transferred between processing stages such as stimulus evaluation and response preparation in a continuous or discrete manner. We tested whether partial information about a stimulus influences the response stage by recording the activity of movement-related neurons in the frontal eye field of macaque monkeys performing a conjunction visual search and a feature visual search with a singleton distractor. While movement-related neurons were activated maximally when the target of the search array was in their movement field, they were also activated for distractors even though a saccade was successfully made to the target outside the movement field. Most importantly, the level of activation depended on the properties of the distractor, with greater activation for distractors that shared a target feature or were the target during the previous session during conjunction search, and for the singleton distractor during feature search. These results support the model of continuous information processing and argue against a strictly discrete model.

Introduction

Measurements of reaction time have played a major role in the development of theories about the cognitive processes that underlie sensation, perception, memory, and action (e.g. [20]). Inferences about information processing from the time needed to perform particular tasks, however, rely explicitly or implicitly on a theory about the temporal relations among the various cognitive processes required to perform those tasks (e.g. stimulus recognition, response preparation and initiation). Thus, not surprisingly, a central issue in mental chronometry has been the nature of transmission of information between stages of the processing system (for reviews, see [9], [15], [16]). Is the transmission of information between stages accomplished in a discrete or a continuous manner?

Discrete information-processing models assume that one process must finish before a subsequent process can begin, so different processes operate in a strictly sequential manner. Such discrete transmission of information between stages has been assumed by the subtraction method [10] and the additive factor method [34], [35], among others.

The assumption that cognitive operations follow one another in strict temporal succession has been criticized though, leading to the development of continuous information-processing models in which a process can transmit partial output before it is completely finished (e.g. [11], [14], [41]). Miller [17] has suggested that an exclusive distinction between discrete and continuous models may be an oversimplification, and that it is possible to construct intermediate models by varying the ‘grain’ size of information that is transmitted. Whereas discrete transmission would be in a single chunk, continuous transmission would be in an infinite number of small grains. From this perspective, the discrete-continuous issue can be recast in terms of whether transmission is ever anything less than fully discrete so that the critical question concerns whether there is a transfer of partial information about a stimulus to the response system before that stimulus has been completely evaluated.

Recently, psychophysiological studies using event-related potentials have addressed the discrete-continuous debate (for review, see [7], [8], [9]). In general, these studies have relied on the P300 potential as a marker for the end of the evaluation process, and the lateralized readiness potential (LRP) as a marker for response preparation and production. The results of these studies have suggested that whether partial information is used may depend on its utility in producing a fast and accurate response.

A series of studies of single neuron activity in primary motor cortex (M1) by Miller, Riehle, Requin and colleagues have found evidence supporting continuous information processing. In an early study, Miller et al. [18] recorded neural activity in M1 of one monkey performing a wrist flexion/extension, go/nogo task. In this task, one stimulus was assigned to a wrist flexion response, and another to a wrist extension response. When a nogo signal was presented, instructing the monkeys to withhold wrist movements, directionally-selective neurons responded with weaker versions of the response patterns to the same visual stimuli when a response was required, suggesting that neurons received partial perceptual information in favor of that movement. Requin and Riehle [22] showed similar results using a left/right, go/nogo task, and obtained additional evidence for continuous transfer of information using a stimulus-response compatibility task in which monkeys aligned a pointer with visual targets on the left or right of a starting position. In the spatially-compatible trials, they had to point at the stimulus location, whereas in the spatially-incompatible trials, they had to point at the target located in the opposite side. Results of this experiment provided evidence of transmission of information in the incompatible trials about the congruent, but incorrect response, before the incongruent, but correct response was programmed. A subsequent study by Riehle et al. [23] extended these findings using a stimulus-incompatibility task, describing neurons in M1 sensitive to the stimulus-response mapping rule, with a large functional and temporal overlap between this population of neurons, and populations sensitive to the stimulus or response side.

In electrophysiological studies described above, the continuous-discrete question was addressed in the motor cortex using tasks in which the response to particular stimuli changed between conditions. We have used a new and different approach to address the same question in the visuomotor system by recording single neuron activity in the frontal eye field (FEF) of macaque monkeys performing two conceptually different visual search tasks commonly employed in human behavioral studies [39], [40], [42].

FEF, located on the rostral bank of the arcuate sulcus in the frontal cortex, plays a key role in transforming the outcome of visual selection into a command to move the eyes (for review, see [26], [28]). Consistent with this role, FEF has both visual and motor characteristics. Roughly, half of the neurons in FEF have visual responses [4], [19], [25], mediated by massive converging input from extrastriate visual areas of both the dorsal (or ‘where’) and ventral (or ‘what’) streams [1], [30]. In previous studies we have shown that visual responses of FEF neurons do not distinguish the target from distractors during either popout [27], [29] or conjunction visual search [2]. However, the activity of visually-responsive FEF neurons evolves to discriminate the target as reflected by a relative suppression of distractor-evoked activity. Furthermore, this selection process is dissociated from saccade production as we have shown that the time at which these neurons discriminate target from distractors does not predict the variability of saccadic reaction times [38], and that this selection does not depend on saccade production or programming [37].

The universally accepted motor function of FEF is mediated by layer 5 movement-related neurons that exhibit little or no sensory response to stimulus presentation but are active specifically before and during saccades [4], [13], and project to the superior colliculus [33], as well as parts of the brainstem saccade-generating circuit [32]. In fact, FEF is defined as the region of frontal cortex from which saccades are elicited with currents of less than 50 μA [5]. Furthermore, unlike visually-responsive neurons, characteristics of the activity of movement-related neurons account for the variability observed in reaction times as described by Hanes and Schall [13].

We have previously investigated saccade target selection in FEF during a visual search in which monkeys shifted gaze to a target defined by the conjunction of color and shape [2]. We found that the activity of many visually-responsive neurons not only discriminated target from distractors, but also discriminated among the distractors based on their visual similarity to the target and based on the history of target properties used across sessions. To the extent that movement-related neurons in FEF represent a processing stage closer to saccade production, discrete and continuous processing models make different predictions regarding the activity of these neurons.

A strictly discrete processing scheme would predict that while visually-responsive neurons participate in discriminating target from distractors, only the final outcome of the discrimination (i.e. target location) would be conveyed to the movement-related neurons. Thus, movement-related neurons would only activate for the saccade that is produced based on the outcome of processing in the visual selection stage. On the other hand, a continuous processing scheme would predict that information about the likelihood that each stimulus, including distractors, is the target of the search influences the activation of the movement-related neurons. Thus, evidence that movement-related neurons are modulated by similarity and priming would support a continuous processing model. Note that this approach does not rely on attributing differential behavioral responses to stimuli across conditions. The task is always to find and shift gaze to the defined target. Instead, we use differences in the visual properties and behavioral significance of the various distractors to determine whether information about these distractors is reflected in the motor preparation and execution stage.

The nature of information transmission between stages was also tested in a modified popout visual search which, unlike conjunction search that relies entirely on a memory representation of the target, is based on conspicuousness. In this task, monkeys searched for the stimulus with the oddball shape. However, we changed the behavioral significance of one of the distractors by changing its color. Studies with human subjects have shown that despite being irrelevant to the task, the color singleton attracts attention (e.g. [36]). We took advantage of this fact and investigated whether movement-related neurons are modulated by the conspicuousness of the distractors. Finding that movement-related neurons are modulated by distractor properties would show that these neurons receive information about more than just the final outcome of visual selection (i.e. target location), and thus would support a continuous processing model.