The hierarchical organization of decision making in the primate prefrontal cortex

Abstract

The prefrontal cortex plays an important role in making the association between sensory information and specific behavior. For example, in a complex stimulus–response situation, such as the Wisconsin card sorting test, prefrontal patients show difficulty in making appropriate decisions. To understand the neural mechanisms, we recorded prefrontal cell activity while monkeys performed a go/no-go selective attention task where the subjects made a go or no-go response depending on the color or the motion direction of compound visual stimuli (moving colored dots). Groups of cells showed differential activity for go and no-go stimuli (go/no-go activity): some showed the activity either in the color or motion attending condition, and others showed the activity both in the color and motion conditions. Cells of shorter latencies, found mainly in the prefrontal subareas receiving visual input, showed go/no-go activity only when task demands necessitated that the monkeys attended to that cell’s preferred visual dimension. We also found cells with longer latencies in the motor-related periarcuate area that showed go/no-go activity regardless of the dimension attended. These results suggest that subareas in the prefrontal cortex play different roles in associating the sensory information with its behavioral meaning and are hierarchically organized to make appropriate decisions in complex tasks.

Introduction

Decision making can be regarded as a process in which an animal tries to maximize benefit by selecting a response that will lead to a positive result in given circumstances. Animals with higher cognitive abilities are able to change their response to the same physical stimulus depending on the context, and so are able to adapt to new stimulus/reward associations. Such higher cognitive function, or complex decision-making process seems to be closely related to the working of the prefrontal cortex (PFC). Neuropsychological studies have implicated PFC in decision making, particularly in polysemous stimulus circumstances. For example, prefrontal patients are unable to shift their attention to the appropriate visual dimension in the Wisconsin card sorting test (Milner, 1964) and are unable to choose the strategy that maximizes their score in a gambling game (Bechara et al., 1997). Their deficits are primarily in response selection and strategy because prefrontal patients can perceive their surroundings normally and do not have difficulty in motor execution (Bechara et al., 1997, Fuster, 1997).

To understand the underlying neural mechanisms, single unit studies have been conducted with non-human primates using various discrimination tasks (Niki, 1974, Watanabe, 1986, Yamatani et al., 1990, Sakagami and Niki, 1994a, Bichot et al., 1996, Miller et al., 1996, Fuster, 1997). These studies have suggested that cells in primate PFC code the representation of the response which should be made to the cue stimulus to obtain reward, given that their activity during the cue period predicted the monkey’s impending action. The activity pattern correlates with the behavior to be executed later but does not have a direct causal relationship with it at the time of execution. The representation is characterized best as the behavioral meaning or significance of the cue stimulus (Watanabe, 1986, Yajeya et al., 1988, Sakagami and Niki, 1994a). For example, in Sakagami and Niki (1994a), monkeys were trained to make a go or no-go response depending on the physical features of the cue stimulus, e.g. a red or yellow color indicated that the monkey should make a go response to obtain reward, and a green or purple color indicated that the monkey should make a no-go response. Many cells in the dorsolateral prefrontal cortex (DLPFC) showed differential visual responses to the cue stimulus, not depending on the physical features, but on whether the feature of the cue stimulus indicated a go response or a no-go response. The differential activity of these cells was not simply related to the motor execution of a go or no-go response as there was a delay period between the cue presentation and the response.

Cells that code behavioral meaning can provide insights into how the PFC plays its important role in making and/or keeping an association between sensory information and a specific behavior. In a complex decision-making situation, however, the appropriate response to a given stimulus may be different depending on the context. How does the PFC make the appropriate association between the stimulus and its behavioral meaning in such a complex situation? Anatomical studies have suggested that the various interconnections of the PFC with sensory cortices and the close intrinsic connections within the PFC could underlie such a higher cognitive function. For example, the inferior convexity (ventral part of Walker’s area 46 and area 12) of the PFC receives color and shape information from the inferotemporal cortex (Barbas, 1988, Seltzer and Pandya, 1989, Ungerleider et al., 1989). On the other hand, the parietal cortex supplies spatial or movement information about a visual stimulus to the principalis (area 46) and arcuate (area 8) areas in the PFC (Cavada and Goldman-Rakic, 1989, Seltzer and Pandya, 1989, Andersen et al., 1990). The processed information in these PFC subareas flows to the periarcuate area (area 8 and sulcal area 6) (Barbas, 1988, Barbas and Pandya, 1989). Thus, we hypothesized that cells in each subarea of the PFC code different aspects of the behavioral meaning of the stimulus, depending on the anatomical connections, and furthermore we posited that the PFC is hierarchically organized to support decision making about which is the appropriate response in a given situation. Such a decision would then be a precursor to the motor command.

We recorded single-unit activity in three male Japanese monkeys while they performed a go/no-go selective attention task in which they had to discriminate compound visual stimuli (targets) consisting of moving colored dots. To test the hypothesis described above, we compared the activity patterns of cells in each PFC subarea in different attending conditions (color or motion), examining response latency and receptive field properties. If our hypothesis based on anatomical connections is true, we should find cells in the inferior PFC that show differential go/no-go activity to the cue stimulus only when the monkey attends to the color of the stimulus, while cells in the superior PFC should show go/no-go activity only when the monkey attends to the motion, and in the periarcuate area we should find cells that show go/no-go activity irrespective of the attending condition. This hypothesis also predicts that condition-independent go/no-go cells in the peri-arcuate area have longer latencies to the cue stimulus and larger receptive fields than condition-dependent go/no-go cells. This is because go/no-go cells that are free from constraints by the visual information processes, and so are closer to cells that generate motor commands, should have a position in the later stage of the hierarchical decision-making process.