Specialization and synergism of prefrontal pathways for cognition, emotion and action
Helen Barbas
Boston University

The primate prefrontal cortex holds a privileged position within the nervous system for cognition, emotion and action. To guide behavior, the prefrontal cortex must select relevant information, disregard irrelevant information, and access motor control systems for action. How are such complex functions coordinated? There is evidence that highly organized pathways link distinct prefrontal sectors with excitatory and inhibitory systems underlying selection of diverse but relevant signals for the task at hand. Lateral prefrontal cortices in rhesus monkeys are robustly linked with occipital, temporal, parietal, and premotor areas, implicated in sensory perception, cognition, working memory and action. Medial prefrontal cortices are richly connected with the hippocampal formation, perirhinal cortices, the amygdala, and central autonomic structures associated with long-term memory, emotional memory, and emotional expression. Caudal orbitofrontal areas receive an overview of the external sensory environment and target dual systems in the amygdala, poised to either drive or dampen autonomic activity that may depend on emotional context. The different sectors of the prefrontal cortex communicate with each other, inextricably linking pathways associated with thoughts and emotions that guide actions. This communication is transmitted through laminar-specific pathways that suggest the sequence of recruitment of structures associated with memory, and sensory and emotional perception. Disconnection of these pathways disrupts normal behavior, as seen in psychiatric diseases affecting preferentially distinct prefrontal cortices. (Supported by NIH grants from NIMH and NINDS.)
Temporal and hierarchical dimensions of executive control in the human prefrontal cortex
Etienne Koechlin
Pierre et Marie Curie University, Paris)

{The prefrontal cortex subserves executive control, i.e. the organization of action or thought in relation to internal goals. In this talk, I will present recent fMRI results from our group showing that the lateral prefrontal cortex is organized as two distinct cascading systems implemented the temporal and hierarchical dimensions of executive control. The first system extends from posterior to anterior and polar prefrontal regions and is involved in processing the various cross-temporal contingencies underlying executive control, thereby implementing the three basic temporal forms of executive control: synchronic or contextual control, diachronic or episodic control, and polychronic or branching control, respectively. The second system is confined to Broca's area and its right homologue, extending from premotor regions to BA 44 and BA 45, and is involved in controling start- and end-states and the nesting of functional segments that form the hierarchical structures of action plans ranging from single motor acts to simple and superordinate action chunks respectively. This model highlights a basic functional segregation between cross-temporal and cross-hierarchical processing in executive control and may partly explain the critical contribution of prefrontal regions to human language.
Abstract state-based inference in human ventromedial prefrontal cortex during reward-based decision making
John O'Doherty
Calitech

Adaptive reward-based decision making in an uncertain environment requires the ability to make predictions about the expected future reward associated with particular sets of actions or stimuli. These predictions are usually learned through experience, and used to guide action selection so that actions associated with greater expected reward are chosen more frequently over the course of learning. Reinforcement learning models (RL) provide a strong theoretical account of how such learning might be implemented. However, these models assume no higher order structure in the decision making problem, such as interdependencies between states, actions, time and ensuing rewards. In this talk I will describe evidence that during performance of a simple decision task with a rudimentary higher order structure, human subjects engage in abstract-state based decision making in which knowledge of the underlying structure of the task is used to guide behavioral decisions rather than standard RL. Moreover, neural activity in human ventromedial prefrontal cortex as measured with fMRI during performance of this task is strikingly consistent with abstract state-based inference and not with standard RL. These results show that a region of the human brain - the prefrontal cortex, employs an abstract predictive model of task structure in order to guide behavioral decision making. This capacity could underlie the ability of humans to predict the behavior of others in complex social transactions and economic games, and accounts more generally for the human ability of abstract strategizing.
The Role of Fluctuations in Decision-Making
Gustavo Deco
Barcelona

We analyze the neurodynamical mechanisms engaged in the process of comparison in a decision-making paradigm from the perspective of Weber's law, that is, we investigate the probabilistic behavior of the neural responses responsible for detecting a just noticeable stimulus difference. An ideal paradigm for studying this is the vibrotactile sequential discrimination task (Romo et al., 2004). In this two-alternative, forced-choice task, subjects must decide which of two mechanical vibrations applied sequentially to their fingertips has the higher frequency of vibration. In particular, single neuron recordings in the ventral premotor cortex (VPC) reveal neurons whose firing rate was dependent only on the difference between the two applied frequencies, the sign of that difference being the determining factor for correct task performance. These neurons reflect the decision-making step of the comparison, and therefore we model here their probabilistic dynamical behavior as reported by the experimental work; and through the theoretical analyses we will relate these neurons' behavior to Weber's law. We analyze and model the activity of these VPC neurons by means of a theoretical framework first proposed by Wang (2002), but investigating now the role of finite-size fluctuations in the probabilistic behaviour of the decision-making neurodynamics, and especially the neural encoding of Weber's law. We will also discuss, the role of the noise induced changes in the probabilistic behavior of the system by applying a semi-analytical approach that applies the moments method to the system of stochastic equations in order to derive a reduced deterministic system of differential equations which describes the moments (first and second order) of the state variables.
Contrasting the roles of the medial and lateral prefrontal cortices in decision-making
Matthew Rushworth
Oxford

Decision-making depends on the integrated activity of several frontal areas but recent neuroimaging and lesions experiments suggest distinct specialization for ventral prefrontal (PFv) and the anterior cingulate cortex (ACC). In the macaque connexions conveying information about the identity of stimuli and possible response selection instructions run from the temporal lobe in the uncinate fascicle and extreme capsule and terminate predominantly in PFv and orbital prefrontal regions. Diffusion weighted imaging and probabilistic tractography methods suggest a similar pattern of connectivity in the human. Neuroimaging experiments reveal that human PFv is activated when subjects use learned instructions to decide which action to make. The ability to use learned rules in order to select a response is compromised after PFv lesions. A critical aspect of the deficit is the inability to select the stimulus information that is relevant for guiding the decision. The ACC has connexions with amygdala and striatal areas, which convey information about reward and motivational value, and with the motor system. ACC lesions disrupt the ability to select actions on the basis of recent reward history. The ACC is activated in neuroimaging studies when subjects generate hypotheses and make decisions about which action will lead to the best outcome.
Neural correlates of executive control in the avian "prefrontal cortex"
Michael Colombo
Dunedin

Executive control, the ability to plan one's behavior to achieve a goal, is a hallmark of frontal lobe function in humans. In the current study we report neural correlates of executive control in the avian nidopallium caudolaterale, a region analogous to the mammalian prefrontal cortex. Pigeons performed a working memory task in which cues instructed them whether stimuli should be remembered or forgotten. When instructed to remember, many neurons showed sustained activation throughout the memory period. When instructed to forget, the sustained activation was abolished. Consistent with the neural data, the behavioral data showed that memory performance was high after instructions to remember, and dropped to chance after instructions to forget. Our findings indicate that neurons in the avian "prefrontal cortex" participate in one of the core functions of executive control, the control of what should be rememebered and what should be forgotten.
Scenarios for persistent activity in cortical network models
Nicolas Brunel
Paris

Persistent activity of neurons in associative cortex is widely thought to underlie active working memory. Traditional models explain persistent activity through feedback loops between excitatory neurons. I will present an alternative scenario in which multistability arises because of interactions between inhibitory neurons.
Short- and long-term reward prediction in cortico-basal ganglia loops
Kenji Doya
Okinawa Institute of Science and Technology

Prediction of both immediate and delayed rewards is crucial for appropriate decision making. While reward-predictive neural activities have been found in variety of cortical and sub-cortical areas, their specificity in terms of the time scale of reward evaluation has not been well studies. Thus we performed a number of functional brain imaging experiments while subject performed tasks that required action selection based on prediction of both immediate and future rewards (Tanaka et al., 2004). In conventional block-design analysis, we found stronger activities in lateral prefrontal cortex, premotor cortex, parietal cortex, the striatum, and the lateral cerebellum. Using reinforcement learning model-based analysis, we found ventro-dorsal maps of prediction time scale within the insular cortex and the striatum. Furthermore, under dietary depletion and loading of tryptophan, a precursor of serotonin, the ventral and dorsal striatal activities correlated with short- and long-term reward prediction, respectively, were differentially modulated. Recent diffusion tensor MRI experiments by Lehericy et al. (2004) showed that this part of the striatum (caudate head and anterior putamen) receive topographic projection from the frontal cortex; ventral part from the orbitofrontal cortex and the dorsal part from the lateral prefrontal cortex. These observations suggest that the parallel circuits linking the frontal cortex and the anterior striatum have a ventro-dorsal specificity in the time scale of reward prediction, and that they are under differential regulation by serotonin, possibly due to differential expression of serotonin receptor subtypes (Compan et al., 1998). Comparison of maps under tasks with different time steps indicates that the map is based on the steps for prediction, rather than length of time. Activation of the lateral prefrontal cortex may reflect the use of forward models of the state dynamics, through its loop connection with the lateral cerebellum.
Reinforcement learning and decision making in prefrontal cortex
Daeyeol Lee
Rochester

To make optimal choices in a dynamic environment, decision makers must evaluate outcomes from previous actions and update their estimates for outcomes expected from future actions. This suggests that signals related to the animal?s previous choices and their outcomes must be integrated across multiple actions. To understand the role of prefrontal cortex in such adaptive decision making, we trained monkeys to perform an oculomotor free-choice task modeled after a competitive game, and examined the activity of neurons in the primate prefrontal cortex (PFC). We found that PFC neurons often modulated their activity according to the previous choices of the animal and computer opponent as well as the rewards in preceding trials. In particular, neurons in the anterior cingulate cortex mostly encoded signals related to the rewards in previous trials, and these signals were often biphasic, indicating that they may be involved in the evaluation of choice outcome by comparing it to a reference point. These results suggest that the prefrontal cortex makes an important contribution to adaptive decision making by combining multiple types of signals related to the animal's previous choices.
Cooperation between prefrontal and visual cortex during working memory
Gregor Rainer
Max-Planck-Institute, Tuebingen

Working memory has been associated with delay activity - increased spiking by neurons in the prefrontal cortex (PFC). PFC neurons showing delay activity often show accelerating firing rates towards the ends of delay periods, and this activity has been shown to be correlated with anticipation of expected stimuli, actions or rewards. We have recently studied activity in extrastriate area V4 during delay periods. V4 neurons do not show general increases in firing rate during delays. However, we have observed an elevation in theta frequency oscillations in the V4 local field potentials. These increase in oscillations had a systematic effect on spike timing during the delay period, such that many neurons tended to fire preferentially at certain phases of the theta cycle. We found that when taking theta phase into account, the majority of neurons carried signals about the encoded stimulus whereas only a small minority did so if theta phase was negleted. I will discuss the relation between these two apparently separate mechanisms by which stimuli can be maintained in working memory.
Representation of strategies and goals in the prefrontal cortex
Aldo Genovesio
NIH

Monkeys used abstract response strategies to select one of three spatial goals. On what we termed first-chance trials, when a symbolic visual cue repeated from the previous trial, the monkeys stayed with their previous goal on the current trial (repeat-stay strategy). When the cue changed from the previous trial, the monkeys shifted from their previous goal to one of the two remaining goals (change-shift strategy). One of these two locations would produce a reward, the other would not. Non-reward led to second-chance trials, on which the monkeys shifted from the most recently chosen (and unrewarded) goal to the remaining change-shift goal (lose-shift strategy). We found that prefrontal cortex cells encoded the specific strategy used on both first- and second-chance trials. By exploiting the fact that the monkeys had to maintain the previous goal in memory on the first-chance trials, but had to maintain future-goal information on second-chance trials, we also found that cells encoded either previous goals or future goals, but not both. The first finding suggests prefrontal neurons play a central role in abstract strategies. The second finding indicates that largely separate neural networks encode previous and future goals. A failure to distinguish these two kinds of goals could cause perseveration, compulsive checking, or the omissions that occur in dementia.
A neural model of flexible sensori-motor mapping: learning and forgetting on multiple timescales
Stefano Fusi,1 Wael F. Asaad,2 Earl K. Miller,2 and Xiao-Jing Wang3
1University of Bern, 2MIT, 3Brandeis

What are the neural mechanisms that enable us to associate sensory information to the appropriate action in a given environment, and rapidly remap sensory flow to motor program whenever demanded by changing behavioral context? We investigated this question using a biologically-based cortical microcircuit model of spiking neurons, endowed with reward-dependent synaptic plasticity. We show that our model quantitatively reproduces experimental observations in a visuo-motor association task with behaving monkeys, particularly the animal's learning behavior after association reversal and the correlated changes in prefrontal neuronal activity. Synaptic plasticity on multiple timescales is shown to be crucial to allow the model to quickly forget the old cue-saccade associations, as well as to capture the statistics of correct associations across many blocks of trials. This model suggests a theoretical framework for the flexible sensori-motor mapping at the core of voluntary action in complex behavior.
Slow reverberatory cortical dynamics underlying cognition
Xiao-Jing Wang
Brandeis University

What are the microcircuit properties that enable a cortical area, such as prefrontal cortex, to subserve cognitive functions, as in contrast to early sensory processing? I will address this question using a biophysically-based network model for working memory and decision making. A salient characteristic of the model is slow reverberatory dynamics that gives rise to persistent activity and time integration, possibly mediated by the NMDA receptors at recurrent excitatory synapses. On the other hand, synaptic inhibition leads to stimulus selectivity, winner-take-all competition, and network's ability to filter out behaviorally irrelevant distractors. Computational results will be discussed in relationship with anatomical and physiological properties of the prefrontal cortex.
Neural mechanisms of spatial working memory: contributions of the dorsolateral prefrontal cortex and the orbitofrontal cortex.
Shintaro Funahashi
Kyoto University

Working memory is a mechanism for short-term active maintenance of information as well as for processing maintained information. Working memory is a fundamental mechanism for many cognitive processes including thinking, reasoning, decision making, and language comprehension. Therefore, understanding neural mechanisms of working memory is crucial for understanding neural mechanisms of these cognitive processes. The dorsolateral prefrontal cortex (DLPFC) has been known to play an important role for working memory processes. Neurophysiological studies using non-human primates have revealed that tonic sustained delay-period activity is a neural correlate of the mechanism of temporary maintenance of information and that delay-period activity represents either retrospective (e.g., sensory inputs) or prospective information (e.g., motor outputs), although the majority of DLPFC neurons had delay-period activity representing retrospective information. However, the DLPFC is not only the brain structure related to working memory. The orbitofrontal cortex (OFC) has heavy reciprocal connections with the DLPFC. Therefore, the OFC could also participate in working memory processes. To examine OFC^Òs participation in working memory, we analyzed characteristics of task-related activities while monkeys performed spatial working memory tasks. Similar task-related activities were observed in OFC neurons as DLPFC neurons exhibited. However, most of delay-period activity in the OFC showed gradually increasing activation and omni-directional selectivity. In addition, reward-period activity was more frequently observed among task-related activity in the OFC. These results indicate that, although the OFC participates in the spatial working memory process, the way it participates in this process is different between these two areas, in that the OFC participates more in motivational aspects than the DLPFC.
Synaptic and electrical signaling and dopamine neuromodulation in microcircuits of the monkey dorsolateral prefrontal cortex
Guillermo Gonzalez-Burgos
Pittsburg

Understanding the neurobiological events underlying dynamic processing in neocortical microcircuits involves the technically demanding task of identifying cell types, the pattern of local connectivity between cells, and the functional properties of the connections. This task is probably achieved at its best by using in vitro brain slice preparations. We have developed methods to obtain living brain slices from areas 9 and 46 of the prefrontal cortex of macaque monkeys. Using whole-cell patch clamp techniques we record from neurons identified visually using infrared differential interference contrast videomicroscopy. We focus on the superficial cortical layers, where most of the pyramidal neuron (PNs) that send output to other cortical areas are located. We found that the morphological and electrophysiological properties of PNs are relatively homogeneous. In contrast, multiple populations of interneurons (INs) were identified based on morphology, electrophysiology, and expression of calcium-binding proteins (parvalbumin, calretinin and calbindin). I will discuss data obtained using single- or multiple-cell recordings from PNs and some of the IN types, performed to address properties that may allow the microcircuit to generate and maintain persistent activity or to evaluate the effects of artificially-generated persistent activity on particular components of the microcircuit. Finally, I will discuss data addressing the effects of Dopamine neuromodulation, tested during some of the experiments described above.
Learning and memory in the prefrontal cortex
Min W. Jung, Eun H. Baeg, Minjung Kim, Yun B. Kim,
Suwon

The ability to modify behavioral strategies in accordance with changes in environment is necessary for survival. The prefrontal cortex (PFC) is likely to be engaged in this adaptive process considering that one important function of the PFC is the planning of future behaviors. To test whether the PFC modifies its activities based on past experience, we investigated learning-induced changes in neural activity and synaptic plasticity in the rat PFC. Single neuron recording studies in behaving animals revealed that PFC neural activities change rapidly in parallel with behavior learning. Moreover, correlated spiking among neurons was altered in the process of learning, and long-term potentiation was induced by high-frequency stimulation in sensory cortical projections to the PFC. These results suggest that PFC neural activity changes dynamically in the process of learning, presumably as a consequence of synaptic weight modification in the PFC. In over-trained animals, correlated spiking among neurons remained similar although neuronal activity varied across two different behavioral tasks. These results are fully consistent with the view that multiple behavioral strategies are represented in a distributed and overlapping manner in the PFC neural network. Our study highlights the importance of learning and memory as an essential component of PFC functions.